Machine Learning
Process
By,
Mrs. P. Aileen Chris
AP(SS), CSE, HITS
Introduction to Machine
Learning Process
Machine learning involves building systems that learn
from data.
This process is iterative and includes several key steps
to ensure models perform well and stay relevant.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
How???
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Raw Data Golden Model
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
1. Defining the Problem
Clearly state the real-world problem.
Identify what predictions or insights are needed.
Consider trade-offs: false positives vs. false negatives.
Set success criteria (accuracy, precision, etc.).
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
2. Data Collection and
Preparation
Gather data from relevant sources.
Clean and preprocess: handle missing values, remove
noise.
Perform feature engineering to improve model input
quality.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
3. Model Selection and Training
Choose a model based on the problem and data.
Split data into training, validation, and testing sets.
Train the model by minimizing error on training data.
Use validation data for hyperparameter tuning.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
4. Model Evaluation and Tuning
Evaluate using metrics like accuracy, precision, recall,
F1-score.
Analyze performance and identify improvements.
Tune hyperparameters or try new models if needed.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
5. Deployment and Monitoring
Deploy the model to a production environment.
Monitor performance in real time.
Retrain periodically with new data to maintain
relevance.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
ML –The Process
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Things to know about ML
Machine learning isn't linear.
You may need to revisit previous steps:
Poor performance? Collect more data.
New patterns? Retrain the model.
Model drift? Monitor and update.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Common Pitfalls to Avoid
Using biased or incomplete data.
Overfitting the model to training data.
Ignoring real-world deployment constraints.
Lack of continuous monitoring.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Real-World Example
Example: Spam Email Detection
Problem: Identify spam emails.
Data: Email text, sender, subject.
Model: Naive Bayes classifier.
Evaluation: Accuracy, false positive rate.
Deployment: Filter in email service.
Monitoring: Update with new spam trends.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Tools and Technologies
Languages: Python, R
Libraries: scikit-learn, TensorFlow, PyTorch
Platforms: AWS, GCP, Azure
Tools: Jupyter Notebooks, MLflow, Airflow
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Group Activity Idea
Task: Design a ML system for predicting student performance.
Groups define problem, data needed, and choose a model.
Present approach to class and get feedback.
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
Q&A / Wrap-up
Questions to discuss:
What step do you think is most important?
Have you encountered ML in your daily life?
Summary:
Define ➜ Prepare Data ➜ Train ➜ Evaluate ➜ Deploy
➜ Monitor
By, Mrs. P. Aileen Chris AP(SS), CSE, HITS
