MINI PROJECT WITH SEMINAR

DISEASE DETECTION USING ML/AI

By
RUKHSUN ARA PARVIN
SUBJECT CODE- PGCS-294
Registration No. -201000411210017
Roll No. -10061220016
Department of Computer Science and Engineering
Maulana Abul Kalam Azad University of Technology, West Bengal
Haringhata, Nadia 741249
18th August, 2021

Session- 2020 to 2022

Page 1 of 24 By|Rukhsun Ara Parvin

Table of Content-

1. Abstract--------------------------------------------------------------------------1
2. Keyword--------------------------------------------------------------------------1
3. Introduction----------------------------------------------------------------1 to 3
4. Machine Learning & AI---------------------------------------------------3 to 4
4.1 Machine learning types-------------------------------------------4 to 6
4.2 Machine learning & A.I. History---------------------------------6 to 7
4.3 Domain/ Application----------------------------------------------7 to 8
5. Role of Artificial Intelligence in Health---------------------------------8 to 9
5.1 Medical history on A.I.-------------------------------------------------9
6. Related work--------------------------------------------------------------9 to 10
7. Different types of machine learning technique i.e. Different types of
Machine Learning Algorithm-----------------------------------------10 to 16
8. Important Keywords---------------------------------------------------16 to 17
9. Discussion--------------------------------------------------------------17 to 22
10. Conclusion------------------------------------------------------22 to 23
11. References ------------------------------------------------------23 to 24

Disease Detection using ML/AI:

1. Abstract

Page 2 of 24 By|Rukhsun Ara Parvin

 Now a days’ different type of diseases is the major reason for increasing mortality rate
among us. Manual diagnosis takes a huge time and effort. To reduce the time and effort,
it is now become important to develop automatic diagnosis system, disease detector for
early detection of different types of disease. Data mining technique is the one which
contribute a lot in development of such system. This paper is relative study on various
machine learning models which are Support Vector Machine (SVM), Logistic Regression,
Decision Tree, Naïve Bayes, MLP, Random Forest, K Nearest Neighbour (KNN), XG-Boost
etc. These are done on the dataset taken from UCI repository, Kaggle and many other
Platform. With respect to the result of accuracy, precision, recall, sensitivity, specificity,
false positive rate, and True positive rate the efficiency of each algorithm is measured
and compared. The aim is to analysis of diseases by machine learning classifier for
sufficient decision making in health care. The aim of this paper is explanation of
Machine learning, & different type of classifiers & their comparison.

2. Keyword- Machine Learning, Artificial Intelligence, Diseases.

3. Introduction- There are lot of programming language in computer science. Like

Python, R, MATLAB etc. Among them Python language is the strongest language among
all, which consisting different type of packages, libraries which are already on it. Most
of the models has implemented through python language. This era is a pandemic era.
Nowadays COVID is the most infected diseases globally and we are now in the 2nd phase
of the pandemic and the world is moving towards the 3rd wave of it. So, it is become
important to predict and detect the characteristics and gene segmentation of COVID
otherwise it will maintain and increase its continuous death rate just like 2nd and 1st
phase of the pandemic. SARS-COV2, MERS COV causes corona virus disease. The risk
factor is close contact (within 6 feet), coughs, sneezes, breathes, talks, touch etc.
Different type of machine learning or artificial technique helps to find such kind of
COVID19 symptoms. Early detection is possible only for machine learning.

Cardio vascular disease (CVD) is another disease where ML draw a vital sign for
detecting this. The reason behind this disease due to high blood pressure, diabetes,
extreme level smoking, Over thinking or hypertension, variation of BMI (Body Mass
index), Obesity, Inactivity, High level cholesterol, unhealthy diet, excessive alcohol and
many other reason. Different kind of heart problems like Arrhythmia, Atherosclerosis,
heart defect, CAD (coronary artery disease), Heart infection, Heart attack, Heart failure,
Stroke, Very much chest pain, Excessive rate of heart beat, Herat pressure, Frustration,
fbs, Heart block, and many other. Machine learning can detect it easily by the help of
individual record what they have symptoms, characteristics, attributes like their sex,
height, weight, level of cholesterol & glucose, blood test, blood pressure (systolic,
diastolic), etc.

Breast Cancer is the most often identified cancer and major reason for increasing
mortality rate
among women. Manual diagnosis takes huge time and the lesser amount of systems,
there is
important to develop automatic diagnosis system for early detection of breast cancer.
Data mining technique which contribute a lot in development of such system. For the
classification of Benign and Malignant Tumours, & early detection of breast cancer we
have used classification technique of machine learning where machine is learned from
some past data which can predict the category of new input. Not only breast cancer but
there also blood cancer, skin cancer, lung cancer can also be detected by Machine
learning technique.

For any type of Disease detection and prediction it’s important to test blood. Hence
Blood Disease detection also important. Different types of Machine learning classifier
help to detect disorder of blood. Except this type of Diseases AI or Machine learning
model can detect Diabetes, Plant disease detection, also.
Different type of DDS (Disease detection system) has been generated where information
is inserted into android app. In DDS real time database are used by some pre trained
machine learning algorithm. Here datasets are deployed in firebase.

Page 3 of 24 By|Rukhsun Ara Parvin

4. Machine Learning & AI- When one’s improve behaviour by some past experience
we will say learning has occurred. Machine learning is basically study of computer
algorithm that allow computer programs to automatically improve through experience.
And here experience is called as data. It’s a type of Artificial Intelligence a data analysis
technique which act like human. By different type of Data analysis technique
meaningful information is extracted from data. And fresh data improve various Machine
learning model. Fresh data means outlier, noise free which will be in standard form.
Based on fresh data Machine learning generate different type of model which are useful
for any type of Diseases detection, forecasting, prediction. It can be act as Decision
maker, Detector, any real time problem solver. There is two concepts learner and
reasoned where Learner will build the model and reasoner use this model. And finally a
particular solution is computed. There exist cloud as well as big data which make
Machine learning more smart. We will see how a computer program is differing from the
machine learning-

Figure 1- How a program works

Figure 2- How data feed in to machine learning

Health hazard is the most critical issue in human life. And it is the reason for variable routine
in our daily life. Machine learning techniques has taken care of our life from medical report,
test by early diagnosis. We are able to known what should we do, which exercise we have
needed, what test should require everything can be predicted by Machine learning.

AI is basically when a Machine want to mimic cognitive function of Humans by some programmed
rule or protocol. This AI gives instruction to machine that how they behave in certain situation. The
difference is Machine will never tired’ by doing some work artificially they are more intelligent than
human. Machine has more problem solving approach rather than one human. Basically we want some
system & software in such a way that they mimic the Human behaviour, AI will have accomplished by
studying & learning how a human brain thinks, the way of learning, decide, work, while they solve
some problems in that exact style a machine will do that. Which outcome occurred in this study we
used it as basis of developing intelligent system as well as smart software.

4.1 Machine learning types

Page 4 of 24 By|Rukhsun Ara Parvin

There are four machine learning technique- Machine Learning technique can be categorized
into four main head that is

a. Supervised Learning technique

b. Unsupervised Learning technique
c. Semi-supervised Learning technique
d. Reinforcement Learning technique

Figure 3- Machine learning technique

Supervised Learning is a subset of Machine Learning that enables us to forecast the output
for unknown or future data. It is the job of Machine Learning to discover a function from
labelled data. Labelled data is a dataset that contains both independent and dependent
variables. I have a dataset that has a large number of active variables.
In supervised learning, we utilise data, which may include both the input and the matching
output. That is, we may have an input x and a matching output y for each data instance.
And from this machine learning system, a model will be constructed so that given a new
observation x, it will attempt to determine what the related y is.

Figure- 4
➢ We split our data into training and test sets.
➢ Using the training set, we train our computer.
➢ Then it’s necessary to create a Machine learning model
➢ Now for checking the performance of that model to get accuracy it’s important to test
data
➢ Now we will pass that data and will see the accuracy
Let’s assume input features are {x1 , 𝑥2 , … , 𝑥𝑀 }, y is a target feature. This values are considered
in a training examples for each example. Now also assume a new training example is given
where only input features are given. No target values are here. Then We need to predict the
value for that target features which is unknown to us. If the value of target value is discrete

Page 5 of 24 By|Rukhsun Ara Parvin

in range, then it will be called as classification. If it will be continuous range, then it will be
called as Regression.
If A is dataset that is labelled with samples xi and each of this inputs are labelled with yi
such that A={(x1 , y1 ),…,( xn , yn )}, xi ∈ X, yi ∈ Y. And the relationship is f:X→Y for each sample
yi = 𝑓( xi ) where f= unknown labelling function. For binary classification Y=0,1 and for multi
class classification Y consists of possible K levels.
Unsupervised machine learning is the process of training a computer utilising data that will
not be categorised or labelled and enabling the algorithm to operate on the data without
direction. Without any previous training data, the machine's job is to categorise unsorted
data according to similarities, patterns, and differences.
Unlike supervised learning, this method does not include an instructor. That is, the machine
will get no training. Thus, the machine is limited in that it will be in restricted mode for the
purpose of discovering hidden structures in unlabeled data, which will be accomplished by
us.

Figure- 5 If data has split in two class what properties they will require

In contrast to supervised learning, we lack any preset characteristic. On the basis of the
data that is provided, we will attempt to construct our own class. And will make every effort
to ensure that whatever class is produced has a high intra-class similarity and a low inter-
class similarity. This method is useful for classifying the input data based on its statistical
characteristics.
Reinforcement learning is a subset of Machine Learning in which the learning system
observes its surroundings and learns the optimal behaviour by attempting to maximise some
concept of cumulative.
Agents monitor their surroundings, choose and carry out certain activities,
reward, and are rewarded in turn (or penalties in certain cases).
Over time, the agent develops a strategy or policy (choice of behaviours) that
optimises its benefits.

Figure- 6 Reinforcement learning

Semi-supervised learning is a hybrid method that combines supervised and unsupervised
learning using labelled and unlabeled data. For data-driven supervised learning. We will

Page 6 of 24 By|Rukhsun Ara Parvin

generate some function from the supervised data, and if we also have unlabelled data in
addition to the labelled data, we may attempt to generate a better function.
Machine learning is concerned with developing a model composed of many disciplines and
conditions structured as a probability estimation function. This is a subfield of Artificial
Intelligence known as machine learning.
There are many types of machine learning classifiers that are helpful for identifying diseases,
diagnosing them, predicting their outcomes, and detecting them early. The usage of Deep
Neural Networks is made possible by the presence of GPUs (Graphics processing units) in
machine learning.

4.2 Machine learning & A.I History

Some machine learning classifiers are given based on time period.

year History

1950s Samuels checker-playing program

1956s A.I. was first introduced based on theoretical concept

1960s 1. Neural Network i.e. Rosenblatt’s perception

2. Pattern Recognition

1970s Natural language processing(symbolic)

1980s 1. Decision tree & rule learning

2. Learning
3. Planning
4. Problem solving approach

1981s Back propagation algorithm

1985s Multilayer Perceptron(MLP)

1990s 1. Support vector machine(SVM)

2. Reinforcement Learning
3. Ensemble learning technique
4. Bayes Net Learning

1997s Ada Boost classifier

2000s Kernel version SVM

2001s Random Forest

2005 Neural Network with Deep layer

2006 Deep learning was coined

2009 Self-driving car was built by Google

2010 Deep learning was used commercially as well

2011 Watson wins Jeopardy

Page 7 of 24 By|Rukhsun Ara Parvin

 2014 Human vision surpassed

2015 Machine translation system by neural network

Table 1- Machine Learning & AI History

4.3 Domain/ Application

Diseases detection/ prediction- There are lot of field and application of Machine
Learning in our life. Specifically, the area of Medical perspective. Like as Future
forecasting, sentiment analysis, diagnose diseases, Early detection & prediction of
various diseases like various types of cancer detection like breast cancer, lung cancer,
blood cancer, skin cancer, diabetes, plant diseases detection etc. In this pandemic
situation machine learning helps for handling covid19 by early detecting. This basically
makes healthcare smarter.

✓ Machine learning also detect chronic diseases, Liver disorder, Heart diseases,
Hepatitis, Parkinson’s disease. The symptoms are different from each other.
Machine learning use this symptom as input and detect that particular
diseases earlier. Here different type disease symptoms, Patient records,
different type of lab measurement, pathological test, Blood test, DNA tests,
previous report is used as inputs in Machine learning for diagnose a disease.
And as a result disease has detected. By the help of this records various
datasets have created & from that future patients have cured early.

✓ Not only medical history ML is used but also useful for image recognition,
object detection, Robot control, for natural language processing, Speech
recognition, fraud detection, & many domain & application machine learning
has & AI has.

In Speech Recognition- Automated Speech Recognition- When we search something on Google,

Chrome, Microsoft edge, Siri we just tell it by our voice they can generate output.
Natural language processing- Language use for human communication like English, Hindi, etc.
But in computer system we don’t use Natural language. There are lot of application of NLP like
automated question answer session, Chat bot, Auto correct, prediction, spam detection,
sentiment analysis, machine translation, language synthesis, spam detection etc. AI can help to
process Natural languages with the interaction between computers system and human.
Image recognition- By given some provided data AI model can automatically find image pattern
and help to process that image by gathering and organizing data. Except this in social media,
Healthcare, Financial section, Agriculture, Education, Data security AI plays a significance role.
It can be used for self-driving car, in CD for recommending some essential product.
In social media like Facebook, Instagram, YouTube, Tube Mate, Pandora they are aware which
videos, song we want to next & which we want listen it happens just because of Artificial
Intelligence.
Human activity recognition- Machine learning is useful for recognizing human activity. For
activity capture smartphone, pulsometer, smartphone, Kinect sensor is used. Various machine
learning model like neural network helps to differentiate people’s activity, their gesture,
movement based acceleration of body, acceleration of gravity, body angular speed etc. Which
can help to prevent obesity.
Advance Machine learning on google cloud- Now a day’s advance machine learning model used
in google cloud for their services. This google cloud machine learning has made a huge
revolution in business. Machine learning gives this opportunity to observe data in a new way and
for new perspective. That makes us smarter. In google, they use Machine learning for optimizing

Page 8 of 24 By|Rukhsun Ara Parvin

 their products by the help of Maps (real time traffic). They are able to detect spam on Gmail,
real time face recognition in photos etc.
In Agriculture - In Agriculture, Machine learning is being used for getting more success &
transforming business. Machine learning & AI gives a magical way that farmers can deal with
different types of plant, animals and their specific diseases. There is some awesome machine
learning model can detect plant diseases earlier. For plant species selection, recognition, Quality
of crop, soil & water for plantation ML model can give more accurate result to farmer and makes
their decision more effectiveness.
In computational genetics- Machine learning can deal with large dataset. It has some algorithms
which can make best prediction on data. In bioinformatics Machine learning, AI creates
significance role here. DNA sequencing of different species is also possible for machine learning.
In Cyber security- Machine learning has great impact on cybersecurity. Different types of
vulnerabilities, threats, attacks can be detected by Machine learning model. To make
confidential information more secure & private; AI, ML can handle this thing & identify them.
In financial services- In financial service ML & AI has key role. For customer service section,
Fraud detection in their services are possible just because of ML & AI service Chatbots helps
which in financial services for customer request, customer reviews, question answers section,
bill payments.
Not just this area Machine learning & A.I. has wide range application in many other things like E-
commerce, Manufacturing, AWS, Self-driving car, political campaigns etc. We don’t have any
concept how much ML can help us & will help in future too.

5. Role of Artificial Intelligence in Health-

Artificial intelligence has great impact on health. For diagnosis and detecting any disease AI assign
machine learning. A.I. authorize doctors and medical practitioners for track out disease and health
condition in accurate manner by less amount of cost and errors. Invention of robotics in technology
for medical facilities has become grew up. Now a day the greatest invention of A.I. viz. robotics
creates a wonderful impact for counselling diseases, for physical therapy in rehabilitation in various
laboratory, hospital and research lab in medical field. Also, for operation and surgery purpose robots
are used. CT scan is the most important invention in A.I filed that helps to detecting various cancer
directly viz. lung cancer. Stroke can be detected by CT scan also. Magnetic resonance imaging (MRI),
EEG can access various types of heart disease. Some important boundary on our daily life that A.I.
makes it easier given bellow.

Figure- How robot helps us

5.1 Medical history on A.I.

Year Medical History

Page 9 of 24 By|Rukhsun Ara Parvin

 1950s Various medical thing was invent like X ray,
Artificial Kidney, Cardiac Pacemaker and EEG
1970s An amazing invention replacements of hip
and knee artificially
1960s Analysing of blood, ultrasound, replacement
of heart valve
1980s Surgery of Laser and vascular grafts

6. Related work-
Machine learning has wide range application. Continuous improvement of Machine
learning entire world used this very much. It has high computation power also and in
large datasets also. It’s offer many essential resources as well for various data analysis.
Here lot of research areas which are related with this topic Disease detection using ML.
Heart disease can detect or diagnosis by using hybrid machine learning model. In
(Khourdifi and Bahaj et al.) they predict heart disease by using different type of
Machine learning model. They applied lot of optimization that include PSO which is
combined with ACO. There are several study on hear disease detection. In a research
(Chidambaram T) et al. predict Heart disease using Naïve bayes, AI network, Support
vector machine classifier (SVM), Random forest classifier, and simple regression method.
They found 98.83% prediction result by Decision tree classifier. This paper was about
comparison different type of Machine learning classifier. And Decision tree gave best
result. In a study, (Fahad Kamal Alsheref et al.) different type of machine learning
classification algorithm like Support vector machine (SVM), K-Nearest Neighbour (KNN),
Regression analysis, Decision tree is used for detection of Blood Diseases. (Naresh Kumar
et al.) focuses on automated disease diagnosis. They selected three harmful diseases
which are coronavirus, heart disease, diabetes. Here a android app has been used where
this three disease can be detected by entering data in to that android app by using real
time dataset.

Machine learning model is used for corona virus handling purpose. By the help of
different type of machine learning technique to determine or predict number of death,
number of recovery, affected people, also number of active cases. In china in was
predicted that when covid will end by Machine learning technique. A recently published
paper (Suparna Biswas et al.) A Hybrid Model based on mBA-ANFIS for COVID-19
confirmed case prediction and Forecast. In another study recently published
Mathematical modelling for decision makin of lockdown during covid19 by SIR model
(Suparna Biswas et al.). ES, LR, LASSO, SVM (Rustam et al, 2020) has been used for the
purpose of estimating number of future affected patients. Data set was collected from
GitHub (Wissel et al, 2020).

It’s known that cancer is terrific disease in todays’ life. And Pakistan is one of the country
which has occurrence of Breast cancer. More than 83,000 cases reported. For this this
early detection is important and it is the best effective mode. Machine learning helps to
early detect and gives us the best outcomes. It requires an effective procedure for
discriminate benign tumours from malignant ones. Here lot of related topic exist where
researcher has been done lot of research on this. [Senturk & Kara, 2014] he has proposed
a model for early diagnosis of breast cancer for patients. He used seven different types
of machine learning algorithm for predictions, for prediction process breast cancer
dataset has been collected from UCI machine learning repository. And during the process
of prediction he used Rapidminer 5.0 the data mining tool to apply data mining technique
on chosen algorithm.

For diabetes disease detection (Al-Zebari & Sengur, 2019) used different type of machine
learning technique like Decision tree (DT), Logistic Regression (LR), DA, SVM, k-NN and
ensemble technique. He used matlab for this purpose. 10-fold cross-validation has been
used here. The classification accuracy obtained by individual classifier and compared.
Among Decision tree (DT), Logistic Regression (LR), DA, SVM, k-NN and ensemble The
best accuracy was given by LR method 77.9%. And worst accuracy was given by
Gaussian SVM technique which is 65.5%. Another related work of breast cancer

Page 10 of 24 By|Rukhsun Ara Parvin

 (Senturk et al., 2014) finding the best way for early detecting breast cancer. Dataset was
collected from UCI machine learning dataset. He used SVM, NB, KNN, Decision Tree for
the purpose of early detecting breast cancer. The higher accuracy was given from KNN
classifier 95.15%. And SVM gave 96.40%.

(Cinarer & Emiroglu, 2019) classifying MR brain image characteristics by the outcomes
of tumour classification techniques. For this the machine learning classifiers like RF
(Random Forest), KNN (K-Nearest Neighbour), LDA, SVM has been used for classifying
MR brain image which are n/a, multicentric, multifocal, gliomatosis. Among this
classifier SVM gave highest precision rate compare to other.

Thyroid is also another disease in todays’ scenario. It has many research areas.
(Kousarrizi et al., 2012). By using SVM classifier and UCI machine learning dataset He
got 98.62% classification accuracy. He used two datasets one collected from UCI machine
learning repository and another from Imam Khomeini hospital which was collected by
IDL (Intelligent Device Laboratory) of K.N. Toosi University of Technology.

Parkinson’s disease (Huriharan et al., 2014) was diagnosed by neural network and SVM
(Support Vector machine). He got 100% classification precision result. (Naql et al, 2020)
detect Lung cancer by DL and dataset was taken from LIDC-IDIR (Menge et al, 2018). He
provides automated disease detection analyser. He also provides classification for
promoting radiologists’ diagnosis. By using SVM (Liu et al., 2020) brain stroke has been
diagnosed from 1157 patients. 83.3% accuracy was obtained. Liver disease diagnosis
approach (Durai et al.,) by UCI machine learning dataset. J48, SVM, NB model was used
and obtained 95.04% accuracy. The objective of this research was prediction of higher
score rate for liver disease detection. Using different cancer dataset (Zebedee et al, 2018)
Cancer disease was detected by Convolution neural network (CNN) base on some gene
expression. Classification accuracy was 100%. Another research work is differentiating
the characteristics of Alzheimer’s disease. (Kulkarni and Bairagi, 2017) proposed SVM
model for identifying characteristics for diagnosis of such type of diseases. 96% accuracy
was obtained by this model.

7. Different type of machine learning Technique I.e. Different types of Machine Learning
Algorithm
There are lot of machine learning algorithm which are developed by Machine learning
scientists for detecting any disease, early diagnosis of disease & prediction instead of
manually detecting. Lot of machine learning technique such as Random forest classifier,
K-Nearest-Neighbour, Support vector machine, Gaussian naïve bayes or Bayes theorem,
Decision tree classifier, Linear Regression, Artificial Neural network, Ensemble learning
classifier, Logistic Regression, K-Means clustering, C-Means clustering, Principle
component analysis, Anomaly detection algorithm, J45, C4.5 and C50 model etc. are
used for early detecting disease based on clinical data. Machine learning can use for
distinguish the cancerous cell and Non-cancerous cell. It can distinguish pneumonia. ML
can detect early growth of Tumour in brain as well malignant and benign growth in
Breast, lungs.

Machine Learning Algorithm

Unsupervised Machine learning Algorithm
Supervised Machine learning
Algorithm
Regression Clustering
Classification
Linear Regression K-Means, Fuzzy C-Means, K-Medoids
Support Vector
Machine
SVR Hierarchical
Discriminant
analysis
Ensemble Methods Gaussian Mixture
Naïve Bayes

Page 11 of 24 By|Rukhsun Ara Parvin

 Decision Tree Neural Network
K-Nearest
Neighbour
Neural Network Hidden Markov M
DBSCAN, Agglomerative Clustering

Table 2
7.1 Support Vector Machines (SVM)- Vapnik and Alexey introduced Support vector
machine 1995s. (Berhard Boser et al., 1992) developed Kernel trick on MMH or Maximum-
Mergin-Hyperplanes. (Corinna Cortex et al., 1993) proposed soft margin concept. For
classification and regression purpose SVM analyse data. Main purpose of SVM is to generate
a decision boundary. This line separates data points into two classes. This decision boundary
is called hyperplane. For hyperplane SVM select those points which will be near to that
boundary. This points are called Support Vector. For non-linearly separable data Kernel trick
is used. In Kernel trick inputs are taken as in low dimension and transform it in higher
dimensional space. Which means that non-separable data will be converted to separable
data. Polynomial kernel, Gaussian radial basis function, Gaussian kernel, Laplace RBF
kernel, Hyperbolic tangent kernel, and ANOVA radial basis kernel are all examples of SVM
kernels.

Figure 4
7.2 Logistic Regression- The term "logistic regression" refers to an expansion of
the term "linear regression." Probabilistic value is assigned using logistic regression.
Logistic function is used in Logistic regression for mapping the predicted values into
probabilities. And output is executed through logistic function (Kousarrizi et al.,
2012) Logistic regression use the concept of Threshold value. Based on threshold
value output can either 0 or 1 (Kousarrizi et al., 2012, Abdulazeez, 2018). Logistic
1
function, logistic(η) = 1+ⅇ−η. The Odds ratio concept in Logistic Regression making
p
computation easier. Odds ratio = 1−p
where p= probability that an event occurs and
(1-p)= probability that the event will not occure.
7.3 Linear Regression- Linear regression classifier in machine learning generates a
linear relationship in between dependent & independent variable. Independent
variable can one or more. This model works well in regression but fails in
classification. Linear regression can be represented as y=a+bx+c, where y= Target
Variable (i.e. Dependent Variable), x= (Predictor Variable) Independent Variable, a=
intercept, c=random error. Here x, y are training datasets.

Page 12 of 24 By|Rukhsun Ara Parvin

 Figure 5
7.4 K-NN- K-NN means k-Nearest-Neighbour classifier. It is non-parametric classifier
(1970’s). It is non-parametric in approach as there are no assumption or hypothesis. We have
to select K number of neighbours. Then need to calculate Euclidean distance. And Based on
similarities or nearest neighbours K-NN choose the K nearest neighbours. Need to count all
data points in each category in k neighbours. If there will a new point, it’s need to assign in
that category which has minimum Euclidian distance or has maximum similarities. K-
Nearest-Neighbour is used for-
➢ Pattern Recognition
➢ Statistical estimation

7.5 DTs- This decision tree classifier is based on a tree structure and has a non-linear
function type. It is composed of a root node, an internal node, and a leaf node. This decision
node makes decisions about route selection. And the test node defines the example's class.It
can be performed on the basis of Greedy. This Greedy approach helps to minimizing depth
of tree. It is a non-parametric machine learning classifier (Al-Zebari et al., 2019). That
attribute will provide the best prediction result which has highest information gain. And it
will select as a root node. And for Gini index that attribute will select for split which will have
lowest Gini index and highest reduction impurity. We will prefer smallest DTs always.

Figure 6- Decision Tree

7.6 SVR- SVR is Support Vector Regression. svr is SVM which can support linear regression
as well as Non-Linear regression. Which principle use SVM, SVR use that’s one. But the SVR

Page 13 of 24 By|Rukhsun Ara Parvin

is used for continuous value where SVM for working with classification or discrete value. The
main concept is distinguishing the best fitted line.
7.7 Naïve Bayes- The naive Bayes method is based on Bayes' Theorem.Features which are
classified by Naïve Bayes classifier, all of this features are independent & equal. It is
probabilistic base classifier. Main purpose of this classifier is generating conditional
probability.
w x x
Bayes theorem says, P ( x )= (P (w) P(w)) | P(x). Here P(x) is same for all classes. P (w) P(w))
needs to be maximized. It’s called maximization problem. If a priori probability P(W) is not
known, then it is assumed that all classes are equally likely.
Join probability can be explained as-
P (W, X1, X2,….,Xn)
=P (X1 | X2,….,,Xn, W). P(X2,…,Xn, W)
= P (X1 | X2,….,,Xn, W). (X2 | X3,….,,Xn, W)P(X3,…,Xn, W)
= P (X1 | X2,….,,Xn, W). (X2 | X3,….,,Xn, W)……P(Xn|W). W
As all features are independent so,
P (X1 | X2,…….,Xn, W) = P(X1 | W)
So, P (W | X1, X2,….,Xn) α P (W , X1, X2,….,Xn)
=P (X1 | W). P (X2 | W) , . . . , P (Xn | W). P (W)
= P (W) ∏ni=1 𝑃 (𝑋𝑖 |𝑊) - - - - - - - - - - - - Naïve Bayes formula

7.8 Ensemble learning classifier- It is a Machine learning algorithm which will construct a
set of Machine learning classifier that will help to classify new data point based on vote of
predictions. This Ensemble technique composition of multiple classifier & it will be more
reliable compare to a single classifier. Basically Bayesian averaging is the primary ensemble
technique.

Figure- 7
In case of Max Voting when predictions will perform based on voting. And majority votes are
considered for the final result. In case averaging just average predictions are considered as
final prediction result. In weighted averaging technique different machine learning model are
imposed with various weights demark of each machine learning model for final prediction.
Except this, stacking classifier is also ensemble learning classifier. This staking classifier is
also used for prediction (on the test data) purpose where lot of models are used for making a
new model. Bagging is bootstrapping aggregating. Bagging is a classifier where need to
bootstrapped the data set for making decision. Multiple training datasets sets (Bag) are
created here by random sampling with replacement. And each of bags a single machine

Page 14 of 24 By|Rukhsun Ara Parvin

learning classifier is trained. Boosting classifier is used for correcting errors of previous
model in each model. This Bagging and Boosting are most important classifier that are used
in machine learning.

Bagging classifier Boosting Classifier

Bagging meta- estimator AdaBoost

Random Forest GBM

XGBM

Light GBM

CatBoost

Table- 3

7.9 K- Means & K- Medoids- Both are Clustering algorithm. Main goal of this clustering
algorithm are minimizing the sum squared distance in between data points and cluster
centre in which cluster we want to assign that data points. Number of clusters will already
be known to us (‘K’= Number of cluster). Objective is compute the optimal centroid. In K-
Medoids medoid means a particular point of cluster. Dissimilarities are measured in between
medoid with other data points of that same cluster. And objective is this dissimilarity should
be minimum.
Except this type of cluster there are also DBSCAB (Density- based spatial clustering of
application with noise), Agglomerative clustering, Fuzzy C- Means clustering belong in
Un-Supervised Machine Learning Algorithm.

7.10 Deep Learning- It is a method for unsupervised machine learning. It may have
worked with organised or unstructured data. The concept is same like ML and AI
that it’s build The learning algorithm by mimic the human brain. It is implemented
through the concept of Neural Network. Neural network mimic the concept of
Biological Neuron. This Neurons are said as brain cell.

Figure 8- Biological Neural Network

From this diagram dendrite are used for providing input to a neuron. In
cell body there are a nucleus which perform some function. Then output
will move through axon & it will go on axon terminal and then this neuron
will fire that output towards the next neuron. These two neurons will never
be connected to each other. There will be huge gap between them. This
gap is called Synaptic gap. In this way a BNN works.

Page 15 of 24 By|Rukhsun Ara Parvin

 Figure- 9 Deep Neural Network

In ANN there are several inputs (x1 , x2,…, xn ) like BNN. There are some
randomized taken weights ( w1 , 𝑤2 , . . , 𝑤𝑛). Now this inputs will provide to the
processing element (like cell bode in BNN). Here weights and input will
have summed up by multiplying them. This summation i.e. s=x1 ∗ w1 + x2 ∗
𝑤2 +. . +xn ∗ 𝑤𝑛 . This S is called as a Transfer Function F(S). Then this F(S)
will go through some activation function. This activation function is
nothing but Threshold value. And the fina l output of neuron is dependent
on it. Then it final result will return value 1 If final result is greater than
Threshold value, i.e. neuron will fire at that case otherwise it will generate
0 value i.e. neuron will not fire. There are lot of activation function like
sigmoid activation function, step function as well. In this way AN N works
which mimic the BNN. In ANN if actual output and desired output are not
same that means there are some error. In in Backpropagation algorithm
we will try to minimize that error. We will continue this process until we
get our desired output as our actual output. By the help of
Backpropagation weights will be adjusted. This process called learning.
Deep networks are nothing but neural network with multiple hidden layer.
In this learning every node/ neuron will be interconnected to each other.
There can be multiple hidden layer. Due to present of lot “Hidden Layer”
the concept deep has come. It is not possible in Machine learning.

8. Important Keywords-
Accuracy-
(True Positive or (TP)) + True Negative or (TN))/ (True Positive or (TP))
+ True Negative or (TN)) + (False Positive or (FP)) + (False Negative or
(FN))
Precision-
(True Positive or (TP))/ (True Positive or (TP)) + (False Positive or (FP))
Recall-
(True Positive or (TP))/ (True Positive or (TP)) + False Negative or
(FN))
F1-Measure- (2*Precision*Recall)/ (Precision + Recall)
True Positive (TP)- When we will get targeted result and obtain
result both true.

Page 16 of 24 By|Rukhsun Ara Parvin

 True Negative (TN)- When we will get targeted result and observed
result both are negative or no.
False Positive (FP)- When actual value is negative and predicted
value is positive.
False negative (FN)- When actual value is positive and predicted
value is negative.
Sensitivity- (True Positive(TP))/ (True positive (TP) + False negative
(FN))
Specificity- (True negative)/ (False positive + True negative)
Confusion matrix - The performance of a model is computed based
on it. Dimension should be m X m where m= # target class.

9.Discussion- In this review paper discussion various Machine learning

model used for early diagnosis. Many researchers have been developed
such kind of Detector or predictor machine where diseases are detected
earlier. Due to having some problem on Machine learning most of
researcher used Deep learning algorithm for better achievement of
Accuracy, precision, recall value. It’s noticed that for corona virus
handling (Suparna Biswas et al., Rustam et a., 2020, Wissel et al. ,) used
ML model. For Breast cancer detection and prediction (Senturk & Kara,
2014, Al-Zebari & Sengur., 2019, Sentark et al., 2014,) used LR, DA, SVM,
K-NN, ensemble technique, NB for detection & they successfully go t a good
classification report. E xcept this Khourdifi and Bahaj et al., Chidambaram T et al., Fahad
Kamal Alsheref et al., Naresh Kumar et al.,) used PSO combined with ACO, Naïve
Bayes, AI network, SVM, RFC, Simple Regression method, DTC. And got higher
accuracy. Another researcher used DTC for blood disease detection. Another
researcher (Huriharan et al., 2014) He got successfully 100% classification precision
result by using Neural network and support vector machine. Except this various
researcher who worked in various Machine learning model for detection diseases that
are included in Related work. Except this various researcher who research in
diseases detection that will be discussed below.

❖ Heart Disease- (Chithambaram T, Logesh Kannan N, Gowsalya M) they used

machine learning model K- Nearest Neighbour, Random Classifier, Correlation
and Support vector machines for getting better accuracy for detecting Heart
Diseases. They used datasets with 70K data, 11 features which was collected from
kaggle.com. Data has been pre-processed by Standardscaler. Binary values used
for scaling data. By using K-NN they gained 63.4% score. Not too much better.
By using RFC classifier, they gain 71.4% accuracy which is better than K-NN.
Also for processing data RFC take little time compare to K-NN. 68.4% accuracy
given from DTC classifier. Basically Gini-index was used for prediction. Finally,
72.5% accuracy was given from SVM (Linear kernel), 86.2% by SVM kernel
(Gaussian).
On other side, (O.E. Taylor) gained 98.83% prediction accuracy by using DTC
classifier. The entire paper about comparing different type of ML model like KNN,
Random Classifier, DTC. They have been successfully predicted score for future
reference. That Heart disease may occur & based on that ML model which gave
best accuracy compare to other that particular ML model can be made for the
objective yield checks that an individual having Heart illness or not.

Page 17 of 24 By|Rukhsun Ara Parvin

❖ Breast Cancer- (Habib Dhahri, Eslam AI Maghayreh, Awais Mahmood, Wail
Elkinai, Mohammed Faisal Nagi 2019). For breast cancer diagnosis they chose
dataset from UCI Machine learning repository. Lot of researcher has used this
same dataset for detecting breast cancer. Which consist of total 569 breast cancer
patients (M= 212, B= 357). By Standard scaler classifier data has been pre-
processed. For features selection PCA (Principle component analysis has been
used) for relevant features by randomized SVD. Ensembles model was used for
this diagnosis with 10- fold cross validation. And they successfully detect breast
cancer.
RESEARCHERS’ Chosen model for Result
Breast cancer
diseases detection
Siham A. Decision Tree (J48), J48 NB SMO
Mohammed, Sadeq Naïve Bayes, SMO
Darrab, Salah A. (Sequential Minimal 75.52% 71.67% 69.58%
Noaman, Gunter Optimization, 10-
Saake 2020 and their Fold cross
paper title is Analysis validation on WBC
of Breast Cancer and Breast Cancer
Detection using dataset.
Different Machine
learning technique.

Pre-processing technique is basically if missing values present in dataset then

based on pre-processing technique they are removed. And after pre-processing
technique maximum time it takes good result compare to accuracy before pre-
processing. This is chosen just because of getting good prediction result. There
are lot of pre-processing technique like MinMaxScaler, Standard scaler etc. They
got result different accuracy as they used discretization filter 7 times. For 7-time
discretization filter they obtained at last 98.20% for J48, 76.61% for NB and
95.32% for SMO. Accuracy significantly increased for j48 but not for NB and
SMO.
And then After pre-processing technique they got result for j48= 74.82%, NB=
75.53% and SMO= 72.66%. They got 98.20%, 76.61%, 95.32% for j48, NB, SMO
after 7 times resampling which is significantly good. These all about breast cancer
dataset. They used same technique on WBC dataset also. And they got 99.24%,
99.12% and 99.56% for J48, NB, SMO classifier by using resample filter
technique. Using pre-processing technique, they got 95.91%, 97.37%, 96.78%
for J48, NB, SMO. So, from this paper it is obvious that for obtaining good
prediction result resample filter, pre-processing technique and discretization
played a vital role for disease detection.

❖ Blood Diseases Detection- (Fahad Kamal Alsheref, Wael Hassan Gomma

2019). They used benchmark dataset with 668 records with 28 parameters for
detecting blood disease based on blood test result. They got highest accuracy
result 98.16% by using LogitBoost classifier. Support vector gave lowest
accuracy. In between high and low Random forest gave 97.12%, DT gave 97%,
Regression gave 96.54%, KNN gave 92.97%, Bayesian network gave 92.86%. MLP
contributed 91.80%. They compute metrics such as TP Rate, FP Rate, Precision,
Recall, F-measure, MCC, and ROC curve to determine the correctness of the
model. Additionally, they calculate the PRC Area and the mean absolute error.

Page 18 of 24 By|Rukhsun Ara Parvin

❖ COVID19, Heart disease, diabetes- Fear of COVID19 increase rapidly day
by day. In this situation early detection can control death rate. (Naresh
Kumar et al., 2021) they sketched automated diseases (COVI19, Heart
disease, diabetes) diagnosis machine learning model using Logistic
Regression. In their model a real time dataset has been used where data
are feed to mobile app & based on data detection is possible. For COVID19
prediction relevant features were chosen travel history location, country,
age, sex, symptoms for prediction purpose. It is consisting of 13174 data
points.

For heart disease detection they used 70K data points. Among them relevant
features like gender, age, height, weight, glucose, smoke, alcohol, cholesterol,
systolic blood pressure, diastolic blood pressure has been used for heart disease
detection. From this relevant features systolic blood pressure, diastolic blood
pressure, age, cholesterol is major reason for heart disease according to their heat
map.

For Diabetes detection total 768 data points has been used. Among them relevant
features are pregnancies, Basel metabolic rate, age, blood pressure for prediction
purpose. And this features are reason for diabetes according to their heat map.
Logistic regression has been used & stored on firebase for all dataset.

By their proposed model they successfully detect disease by asking some question
just a few second. In their work, they split their dataset into 3 portion training
(75%), testing (25%) & validation (10%). Accuracy & F- measures for this proposed
model has been predicted which is given bellow-

Diseases Accuracy F- Measures

Covid19 1.4765% 1.2782
Diabetes 1.8274% 1.7264
Heart diseases 1.7362% 1.3821

So, the prediction result is comparatively good. In future we will try feed more
data on android app & will try deployed firebase dataset like this.

❖ Parkinson’s disease
Another disease called Parkinson’s disease for which no particular test is
generated. When nerve cells of brain die Parkinson’s disease occurs. Lot of
symptoms are here like Tremor, changes of speech, writing, Loss of
movements or slower movement, Constipation, Weak muscular etc. Early
detection important. Various researcher works on it by Machine learning
model, Artificial model. Machine learning is the only way which can detect
this disease within few second. (V. Ulagamuthalvi et al., 2020) used
machine learning model for identification of Parkinson’s disease using
XGBoost classifier & Logistic Regression for purpose of classification. They
used UCI dataset which consisting of lot of audio signals. For pre-
processing data Min-max scaler has been used. By XGBoost classifier 96%
accuracy has been occurred where Logistic Regression gave only 79%
accuracy. So, XGBoost classifier works well in Parkinson’s disease

Page 19 of 24 By|Rukhsun Ara Parvin

 identification. Except this more researcher works on Parkinson’s disease
which are given bellow-

RESEARCHERS’ Chosen model for Result

Parkinson’s disease
detection
Mohammad et al., 2014 SVM, Random Tree, 97.37% by FBANN
BPNN (Back Among 195 patients
propagation neural total number of
networks) or FBANN Parkinson’s patients=
(feed forward back 23.
propagation neural
network), 10-fold
cross validation on
195 dataset (voice
samples).
Ramani et al., 2011 Logistic Regression, More than 90%
LDA, Random tree, through LDA &
SVM on UCI dataset Random tree.

Except this disease more no of diseases are detected by ML/AI model which
are given-
Diseases Researcher’s Technique & Result
Chronic Kidney Disease S. Revathy et al., 2019 They used SVM, RFC,
prediction DTs for prediction. After
training and pre-
processing data DT
gave 94.16% accuracy.
Where among 120 #
instances 113
instances were correctly
classified.

SVM gave good result

better than DT was
98.33%. And among
120 instances 118
instances were correctly
classified.

RFC gave best result

compare to them which
is 99.16%. And 119
instances were correctly
classified. So, Random
forest is the best model
for Chronic disease
prediction.

Page 20 of 24 By|Rukhsun Ara Parvin

 They predict the result
based on confusion
matrix.
Diabetes prediction Hasan Tahir Abbas et The main purpose of
al., 2019 paper is defining
demography for
preventing from this
disease. They used SVM
and 10-Fold cross
validation for gaining
better result. And they
got 84.1% validation
accuracy where 81.2%
is recall no to much
better but significantly
good.
Fatty liver disease Cieh-Chen Wu et al., Total number patients
prediction using 20018. are 577. Random
Machine learning Forest, Naïve bayes,
technique. ANN, LR has been used
for prediction. And RFC
gave the best accuracy
result with 87.48%.
This prediction system
can take step for
preventing this disease.
Early diagnosis possible
also. Among 577
patients 377 patients
were correctly detected.
They used 10-Fold
cross validation also for
achieving the better
result. So, RFC is good
classifier for predicting
liver disease.
Thyroid disease Shaik Razia et al., 2017 Actually they gave a
prediction review on Thyroid
disease. Datasets size
72k where features
were 22. There ere no
missing value on
dataset so they got
result. They obtain
99.23% accuracy for
Decision tree. Multiple
linear regression gave
91.59%, SVM gave
96.04% and NB gave
6.31% the worst result

Page 21 of 24 By|Rukhsun Ara Parvin

 given from NB. S, for
Thyroid prediction we
can say DTs is a good
choice.

10. Conclusion- Machine learning has capability of handling large amount of data.
And accuracy, precision and recall depend on the quality of dataset. There are lot of
machine learning techniques helpful for automatic diseases detection. And all
algorithm has different procedure for detection. Different type diseases experiment
by Machine learning model has been conducted for different disease like covid19,
heart disease, different type of cancer etc. So, lot of disease has been discussed here
for the purpose of decreasing high risk from disease. I have discussed about Cancers,
Chronic Kidney disease, Hepatitis, Thyroid, Liver, Blood Disease, COVID19,
Diabetes, Heart disease etc. I have explored and give a review from various
researchers what they did on DDS (Disease detection system).
Different researcher got different and excellent result. It’s necessary all diseases
should detect within limited time. But I have observed in various ML model gave
worse result just because of high dimensional data in which case early detection is
not possible. I have observed (Chithambaram T et al., 2019) they got 63.4% score for
heart diseases detection which is not too good by using K-NN classifier (not too much
relevant features). The dataset was huge. And also KNN took huge amount of time
for processing it. In this review paper I understood every ML classifier that researcher
have chosen worked well & gave good result. It’s shown Decision tree gave best
prediction result 98.83% for Heart disease detection which was good accuracy result
compare to other model. For diabetes disease detection 77.9% was the prediction
accuracy which is moderate not too good. So, it’s need to do more work on it. We will
try to build ensemble model, Deep learning model on diabetes dataset that will give
best classification result. SVM worked well in Breast cancer UCI machine learning
dataset gave accuracy result 96.40%. SVM gave 98.62% result on Thyroid disease
detection which is awesome & amazing. Another disease that for Parkinson’s
diagnosed 100% classification precision result was given by SVM. SVM gave 83.3%
accuracy for brain stroke diagnosed. J48 models gave 95.04% accuracy liver disease
diagnosis. For Alzheimer’s disease researcher got 100% classification accuracy,
another researcher got 96% accuracy by using SVM. So, in nutshell we can conclude
that SVM is a good classification algorithm & smart compare to other which showed
high accuracy. And also it’s noticed that SVM is used by every researcher frequently.
Also I have noticed 10-fold cross validation has been used for every classifier. But for
Parkinson’s disease detection FBANN gave prediction result good which is
significantly good compare to SVM which provided good specificity only that was
100%. The ML algorithm which has some good side and bad side also.

Classifiers Pros. Cons.

Support Vector machine 1. It can be useful for 1. As it is suitable for
linear separable linear separable
dataset as well dataset when
non-linear datasets are in
separable dataset non-separable
as it consists of kernel trick need to
Kernel function. apply. Kernel
2. Chances of convert input
overfitting less. datasets (lower

Page 22 of 24 By|Rukhsun Ara Parvin

 3. I have seen dimensional) into
maximum higher dimensional
researcher used output sets. And
SVM classifier for hence
disease detection computational
and prediction and time will be taken
got max time best as large.
result. Fewer
amount of time
SVM gave worst
result.
Naïve Bayes 1. Those features 1. For training
are not suitable purpose, it
for ML model requires large
can be removed datasets. So, if in
by NB classifier. case of small size
2. It takes less dataset can run on
amount of time. this model result
can come as a bad.
NN Error can be adjusted by Datasets should be large
learning. 1. Lot effort
DTs It requires less effort If It requires lot time for
missing values are training.
present in dataset it will
not effect on DTs during
training.
RFC Random forest consisting It takes huge time for
of decision tree. It can act training.
Ensemble classifier which
can give good result.
Missing values are
automatically handled by
it.
For high dimensional data ML unable to deal with it. And classification report not
too much good if Ml is applied in Big data. As for high dimensional data we need
huge input data as well as output data. Features are unique character that sample
has. It can be considered as variables. In AI these variables are nothing but features.
A big problem of Machine learning is features extraction. Machine learning models
are not efficient for automatically generate features. But the Deep learning will
automatic generate features. Deep learning model are capable of for focus on the
right features. As Deep learning has huge computation power that’s why it can solve
the dimensionality problem that it can deal with high D data. So, hence in future It
should to try detecting diseases by DEEP learning as well.

12. References
[1] (Nareen O.M.Salim & Adnan Mohsin Abdulazeez, 2021). Human Diseases
Detection Bases On Machine Learning Algorithms: A Review.
[2] (Meherwar Fatima & Maruf Pasha, 2017). Survey of Machine Learning for Disease
Diagnostic.
[3] (Shaik Razia, Swathi Prathyusha, Vamsi Krishna, & Sathya Sumana, 2018). A
review on disease diagnosis using machine learning techniques.

Page 23 of 24 By|Rukhsun Ara Parvin

[4] (Md. Mohaimenul Islam, et al., 2018). Prediction of Fatty Liver Disease using
Machine Learning Algorithms.
[5] (K. Subhadra & Vikas Boddu, 2019). Neural Network Based Intelligent System for
Predicting Heart Disease.
[6] (Hasan Tahir Abbas, Lejla Alic, Marelyn Rios, & Muhammad A Abdul-Ghani,
2019). Predicting Diabetes in Healthy Population through Machine Learning.
[7] (S.Revathy, B.Bharathi, P.Jeyanthi, & M.Ramesh, 2019). Chronic Kidney Disease
Prediction using Machine Learning Models.
[8] (Siham A. Mohammed, Sadeq Darrab, Salah A. Noaman, & Gunter Saake,
2020). Analysis of Breast Cancer Detection Using Different Machine Learning
Techniques.
[9] (V. Ulagamuthalvi, G. Kulanthaivel, G. Sri Nikhil Reddy, & G. Venugopal, 2020) .
Identification of Parkinson’s Disease Using Machine Learning Algorithms.
[10] (Naresh Kumar, Nripendra Narayan Das, Deepali Gupta, Kamali Gupta, & Jatin
Bindra, 2021). Efficient Automated Disease Diagnosis Using Machine Learning
Models.
[11] (Fahad Kamal Alsheref & Wael Hassan Gomaa, 2019). Blood Diseases Detecting
using Classical Machine Learning Algorithms.
[12] (Habib Dhahri, Eslam Al Maghayreh, Awais Mahmood, Wail Elkilani, &
Mohammed Faisal Nagi, 2019). Automated Breast Cancer Diagnosis Based on
Machine Learning Algorithms.
[13] (Chithambaram T, Logesh Kannan N, & Gowsalya M). Heart Disease Detection
Using Machine Learning.

Page 24 of 24 By|Rukhsun Ara Parvin