Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Experiment No. 1
Aim: Real estate agents want help to predict the house price for regions in the USA. He gave you the
dataset to work on and you decided to use the Linear Regression Model. Create a model that will help
him to estimate what the house would sell for.

URL for a dataset: https://github.com/huzaifsayed/Linear-Regression-Model-for-House-Price-

Objective:
The main objectives of this experiment are as follows:

1. Understand the fundamental principles and concepts underlying simple and multiple linear
regression techniques.

2. Implement linear regression algorithms to analyze the relationship between house prices and
various independent features such as location, size, number of bedrooms, etc.

3. Evaluate the performance of the linear regression model in accurately predicting house prices
based on the selected features.

Theory:
Linear regression is a widely used statistical method for modeling the relationship between a
dependent variable (target) and one or more independent variables (predictors). In the context of
house price prediction:

Simple Linear Regression:

• Simple linear regression involves a single independent variable (e.g., house size, number of
bedrooms) and a dependent variable (house price).

• The relationship between the independent and dependent variables is modelled using a
straight line equation: Y=β0+β1X, where Y represents the dependent variable (house price), X
represents the independent variable, β0 is the intercept, and β1 is the slope coefficient.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Fig 1: Regression Model

Multiple Linear Regression:

• Multiple linear regression extends the simple linear regression model to incorporate multiple
independent variables.

• The relationship between the dependent variable and multiple independent variables is
modeled using the equation: Y=β0+β1X1+β2X2+...+βnXn, where X1,X2,...,Xn represent the
independent variables, and β0,β1,...,βn are the corresponding coefficients.

The primary objective of linear regression is to find the best-fit line that minimizes the error between
the predicted and actual values, thereby accurately estimating house prices based on the provided
features.

Applications:
Linear regression has numerous applications in real-world scenarios, including:

• Real estate: Predicting house prices based on features such as location, size, amenities, etc.

• Finance: Analyzing the relationship between economic indicators and housing market trends.

• Market research: Estimating sales figures based on advertising expenditure and consumer
demographics.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Input:
Dataset: House Price Prediction

• The dataset contains information on various features of houses (e.g., size, number of
bedrooms, location) and their corresponding selling prices.

Output:
The output of this experiment includes:

1. Predicted house prices: The model generates estimated house prices based on the provided
features using the linear regression algorithm.

2. Evaluation metrics: Metrics such as mean squared error, root mean squared error, and R-
squared value are used to assess the accuracy and performance of the linear regression model.

Conclusion:
Through the application of linear regression techniques, we have successfully developed a predictive
model for estimating house prices in regions across the USA. By analyzing the relationship between
house features and selling prices, the model provides valuable insights for real estate professionals to
make informed decisions.

Outcome:
The outcome of this experiment is a trained linear regression model capable of accurately predicting
house prices based on the selected features. This model can be utilized by real estate agents, property
developers, and prospective buyers/sellers to estimate the market value of residential properties.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Questions:
1. What are the key assumptions underlying the linear regression model?

2. How do you select and preprocess features for inclusion in the regression analysis?

3. What techniques can be employed to address multicollinearity among independent variables?

4. How do you interpret the coefficients obtained from the regression model?

5. What are the implications of heteroscedasticity on the validity of the regression results?
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Experiment No. 2
Aim: Build a Multiclass classifier using the CNN model. Use MNIST or any other suitable dataset.
a. Perform Data Pre-processing

b. Define Model and perform training

c. Evaluate Results using confusion matrix.

Objective:
The main objectives of this experiment are as follows:

1. Develop a deep learning model architecture suitable for image classification tasks.

2. Train the CNN model on the MNIST dataset to learn features and patterns associated with
handwritten digits.

3. Evaluate the performance of the trained model in accurately classifying digit images into their
respective numerical classes.

Theory:
Convolutional Neural Networks (CNNs) are a type of deep learning model commonly used for image
classification tasks. They are characterized by their ability to automatically learn hierarchical
patterns and features from input images. Key components of CNNs include convolutional layers,
pooling layers, and fully connected layers. CNNs excel at capturing spatial hierarchies of features in
images, making them well-suited for tasks like digit recognition in the MNIST dataset.

Applications:
CNNs have a wide range of applications in computer vision tasks, including:

• Object detection

• Facial recognition

• Image segmentation

• Handwriting recognition (as demonstrated in the MNIST dataset)

Input:
Dataset: MNIST dataset

• The MNIST dataset consists of 28x28 pixel grayscale images of handwritten digits (0-9).
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Output:
The output of this experiment includes:

1. Trained CNN model capable of classifying handwritten digits.

2. Confusion matrix to visualize the performance of the model in classifying digits.

Conclusion:
In this analysis, we successfully developed and trained a CNN model to classify handwritten digits
from the MNIST dataset. By evaluating the model's performance, we gained insights into its
accuracy and effectiveness in digit recognition tasks.

Outcome:
The outcome of this experiment is a trained CNN model capable of accurately classifying
handwritten digits. This model can be deployed in various applications requiring digit recognition,
such as optical character recognition (OCR) systems and automated form processing.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Questions:
1. How does the architecture of a CNN facilitate feature extraction and hierarchical learning?

2. What techniques are commonly used for data augmentation in CNN training to improve model
generalization?

3. How do convolutional and pooling layers contribute to the spatial hierarchies of features learned
by the CNN?

4. What are some common challenges or limitations encountered when training CNN models on
image datasets like MNIST?

5. How can transfer learning be applied to CNN models trained on datasets similar to MNIST for
tasks with limited labelled data?
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Experiment No. 3
Aim: Design RNN or its variant including LSTM or GRU
a) Select a suitable time series dataset. Example – predict sentiments based on product reviews

b) Apply for prediction

Objective:
The main objectives of this experiment are as follows:

1. Understand the architecture and functionality of RNNs, LSTM, and GRU networks.

2. Implement an RNN-based model to analyze sequential data and perform sentiment analysis
on product reviews.

3. Evaluate the performance of the designed model in predicting sentiments accurately.

Theory:
Recurrent Neural Network (RNN): RNN is a type of neural network architecture designed to handle
sequential data by feeding the output of the previous step as input to the current step. This
architecture is suitable for tasks where the order of input data matters, such as time series analysis
and natural language processing.

Long Short-Term Memory (LSTM): LSTM networks are a type of RNN that includes specialized
mechanisms to capture long-term dependencies in sequential data. They achieve this by
incorporating memory cells and gates to control the flow of information, allowing them to retain
important information over long sequences.

Gated Recurrent Unit (GRU): GRU is another variant of RNN that simplifies the architecture of
LSTM while achieving comparable performance. It consists of fewer parameters and operations,
making it faster to train and more efficient for certain tasks.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Applications:
RNNs, LSTM, and GRU networks have various applications, including:

• Sentiment analysis of product reviews

• Speech recognition

• Time series forecasting

• Language translation

Input:
Dataset: Time series dataset of product reviews

• The input data consists of sequences of textual product reviews, along with their
corresponding sentiment labels (positive, negative, neutral).

Output:
LSTM Architecture GRU Architecture
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Conclusion:
In this analysis, we successfully designed and implemented an RNN-based model, including LSTM
or GRU variants, for sentiment analysis of product reviews. The performance of the model was
evaluated using various metrics, demonstrating its effectiveness in predicting sentiments accurately.

Outcome:
The outcome of this experiment is a trained RNN model capable of accurately predicting sentiments
based on product reviews. This model can be deployed in real-world applications to analyze
customer feedback and sentiment trends.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Questions:
1. How do LSTM and GRU networks address the problem of vanishing gradients in traditional
RNNs?

2. What are the advantages of using RNNs over traditional machine learning algorithms for
sequential data analysis?

3. How can the choice of hyperparameters impact the performance of an RNN model in
sentiment analysis tasks?

4. What preprocessing techniques are commonly applied to text data before feeding it into an
RNN model?

5. How can techniques like dropout and regularization be used to prevent overfitting in RNNs?
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Experiment No. 4
Aim: Design and implement a CNN for Image Classification
a) Select a suitable image classification dataset (medical imaging, agricultural, etc.).

b) Optimized with different hyper-parameters including learning rate, filter size, no. of layers,
optimizers, dropouts, etc.

Objective:
The main objectives of this experiment are as follows:

1. Develop a CNN architecture capable of accurately classifying handwritten digits from the
MNIST dataset.

2. Optimize hyperparameters to improve the CNN model's performance in terms of accuracy and
efficiency.

3. Evaluate the effectiveness of different hyperparameter configurations in enhancing model

Theory:
Convolutional Neural Networks (CNNs) are deep learning architectures specifically designed for
image classification tasks. They consist of several layers, including convolutional layers, pooling
layers, and fully connected layers.

• Convolutional Layers: Extract features from input images through convolution operations.

• Pooling Layers: Downsample feature maps, reducing spatial dimensions while preserving
important information.

• Fully Connected Layers: Perform classification based on learned features extracted by

Analysis Steps:
1. Data Preprocessing:

• Load the MNIST dataset and normalize pixel values to the range [0, 1].

2. Model Architecture:

• Design a CNN architecture with convolutional, pooling, and fully connected layers,
incorporating dropout for regularization.

3. Hyperparameter Optimization:

• Experiment with different hyperparameters, including learning rate, filter size, number of
layers, optimizers (e.g., Adam, SGD), and dropout rates, using techniques like grid search or
random search.

4. Training:

• Train the CNN model on the training set and validate it on a validation set.

5. Evaluation:

• Evaluate the model's performance on a test set using metrics such as accuracy, precision,
recall, and F1-score.

6. Analysis:

• Analyze the impact of different hyperparameters on the model's performance and identify the
optimal configuration.

7. Conclusion:
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

• Summarize the findings and discuss the implications of hyperparameter optimization on

Input:
The input for this experiment is the MNIST dataset, which comprises grayscale images of
handwritten digits (0-9). Each image is represented as a 28x28 pixel matrix, with pixel values
ranging from 0 to 255. The dataset is divided into training, validation, and test sets, containing
labeled images for model training, validation, and evaluation, respectively.

Output:
The output of this experiment includes:

1. Trained CNN model capable of classifying handwritten digits.

2. Confusion matrix to visualize the performance of the model in classifying digits.

Conclusion:
Through systematic experimentation and hyperparameter optimization, we successfully designed and
implemented a Convolutional Neural Network (CNN) for image classification using the MNIST
dataset. By fine-tuning various hyperparameters such as learning rate, filter size, number of layers,
optimizers, and dropout rates, we significantly improved the model's performance in accurately
classifying handwritten digits. This experiment highlights the importance of hyperparameter
optimization in achieving optimal results in deep learning tasks, particularly in image classification.

Outcome:
Through systematic experimentation and hyperparameter optimization, we successfully designed and
implemented a Convolutional Neural Network (CNN) for image classification using the MNIST
dataset. By fine-tuning various hyperparameters such as learning rate, filter size, number of layers,
optimizers, and dropout rates, we significantly improved the model's performance in accurately
classifying handwritten digits. This experiment highlights the importance of hyperparameter
optimization in achieving optimal results in deep learning tasks, particularly in image classification.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Questions:
1. How do different hyperparameters, such as learning rate and filter size, influence the
performance of a CNN model in image classification tasks?

2. What strategies can be employed to effectively optimize hyperparameters and improve the
performance of CNN models?

3. How does the choice of optimizer affect the convergence speed and final accuracy of a CNN
model?

4. What are the advantages and disadvantages of using dropout regularization in CNN
architectures?

5. How can techniques like grid search and random search be utilized to efficiently explore the
hyperparameter space and identify the optimal configuration?
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Experiment No. 5
Aim: Perform Sentiment Analysis in the network graph using RNN.

Objective:
The main objectives of this experiment are as follows:

1. Develop a sentiment analysis model using RNNs to classify text into positive, negative, or neutral
sentiments.

2. Understand the theory behind RNNs and their application in sequential data analysis, particularly
in sentiment analysis tasks.

3. Evaluate the performance of the RNN-based sentiment analysis model using appropriate evaluation
metrics.

4. Analyze the results to identify strengths, weaknesses, and potential areas for improvement in the
model.

Theory:
1. Recurrent Neural Networks (RNNs): RNNs are a type of neural network designed to work with
sequential data. They have connections between nodes forming directed cycles, allowing them to
maintain a memory of previous inputs. This makes them effective for tasks involving sequences
such as text analysis.

2. Sentiment Analysis: Sentiment analysis involves classifying text into different categories such as
positive, negative, or neutral based on the sentiment expressed in the text. It is widely used in
various applications such as customer feedback analysis, social media monitoring, and opinion
mining.

Recurrent Neural Network

Applications:
The application of sentiment analysis using RNNs is widespread across various domains, including:

• Social media monitoring: Analyzing sentiment in social media posts and comments to understand
public opinion.

• Customer feedback analysis: Classifying customer reviews into positive, negative, or neutral
sentiments to gather insights into product satisfaction.

• Opinion mining: Extracting sentiment from news articles, blogs, and other textual sources to gauge
public sentiment on specific topics.

Input:
Textual data containing sentences or paragraphs for sentiment analysis.

Output:
Classification of text into positive, negative, or neutral sentiments based on the sentiment expressed in
the input text. Below is the confusion matrix showing the results.

Conclusion:
In this experiment, we successfully developed and evaluated a sentiment analysis model using
Recurrent Neural Networks (RNNs). The RNN-based model demonstrated the ability to classify text
into different sentiment categories effectively. The analysis highlighted the importance of RNNs in
capturing sequential information and their utility in sentiment analysis tasks. Further experimentation
and fine-tuning could potentially enhance the model's performance in sentiment classification tasks.

Outcome:
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

The outcome of this experiment is a trained sentiment analysis model using RNNs, capable of
accurately classifying text into positive, negative, or neutral sentiments. This model can be deployed
in various applications requiring sentiment analysis to gain insights from textual data.
Subject Code: 417532(D) Subject Name: Deep Learning SEM-II

Questions:
1. How does the architecture of RNNs enable them to handle sequential data effectively?

2. What are some common challenges encountered in sentiment analysis tasks, and how can
RNNs help address these challenges?

3. How do different hyperparameters impact the performance of RNN-based sentiment analysis

4. What preprocessing techniques are typically applied to text data before feeding it into an
RNN model for sentiment analysis?

5. How can techniques like attention mechanisms be integrated into RNN-based sentiment
analysis models to improve performance?