ML Model and Prediction Flow
The machine learning model flow describes how data is processed, a model is trained, and predictions
are made.
1. Data Collection
• Gather raw data from sources such as databases, sensors, surveys, or logs.
• Example: student marks, customer data, images, text, etc.
2. Data Preprocessing
• Clean the data by removing noise, missing values, and duplicates.
• Convert data into a suitable format.
• Steps include:
o Data cleaning
o Data normalization
o Encoding categorical values
3. Feature Selection / Feature Engineering
• Select important features (input variables).
• Create new features if needed.
• Helps improve model accuracy and reduce complexity.
4. Train–Test Split
• Split data into:
o Training data → to train the model
o Testing data → to evaluate the model
• Common ratio: 70:30 or 80:20
5. Model Selection
• Choose an appropriate ML algorithm.
• Examples:
o Linear Regression
o Decision Tree
o K-NN
o SVM
o Neural Networks
6. Model Training
• The algorithm learns patterns from training data.
• Model parameters are adjusted to minimize error.
7. Model Evaluation
• Test the model using unseen test data.
• Common evaluation metrics:
o Accuracy
o Precision
o Recall
o RMSE (for regression)
8. Model Optimization (Optional)
• Improve performance using:
o Hyperparameter tuning
o Cross-validation
9. Prediction
• New input data is given to the trained model.
• The model outputs predictions.
• Example: class label, price, probability, etc.
10. Deployment
• Deploy the trained model into real-world applications.
• Example: web app, mobile app, or API.
11. Monitoring and Updating
• Monitor performance over time.
• Retrain model when new data is available.
Simple Flow Diagram
Data Collection
Data Preprocessing
Feature Selection
Train-Test Split
Model Selection
Model Training
Model Evaluation
Prediction
Deployment
ML Pipeline – Overview
An ML pipeline is a sequence of steps that automates the process of building, training, evaluating, and
deploying a machine learning model.
1. Problem Definition
• Clearly define the objective.
• Decide whether it is:
o Classification
o Regression
o Clustering, etc.
2. Data Collection
• Collect data from multiple sources:
o Databases
o Sensors
o APIs
o Logs
3. Data Preprocessing
• Prepare raw data for learning.
• Includes:
o Handling missing values
o Removing noise & outliers
o Normalization / standardization
o Encoding categorical data
4. Exploratory Data Analysis (EDA)
• Understand data patterns and relationships.
• Use statistics and visualizations.
5. Feature Engineering
• Select relevant features.
• Create new meaningful features.
• Reduce dimensionality if needed.
6. Data Splitting
• Divide data into:
o Training set
o Validation set
o Test set
7. Model Selection
• Choose suitable ML algorithm.
• Examples:
o Linear Regression
o Decision Trees
o SVM
o Neural Networks
8. Model Training
• Train the model using training data.
• Learn parameters and patterns.
9. Model Evaluation
• Evaluate model performance.
• Metrics:
o Accuracy
o Precision / Recall
o F1-score
o RMSE
10. Hyperparameter Tuning
• Optimize model settings.
• Techniques:
o Grid Search
o Cross-Validation
11. Prediction
• Use trained model to predict unseen data.
12. Deployment
• Integrate model into production systems.
• Example:
o Web app
o API
o Mobile app
13. Monitoring & Maintenance
• Track model performance.
• Retrain model when data changes.
ML Pipeline Flow Diagram
Problem Definition
Data Collection
Data Preprocessing
EDA
Feature Engineering
Data Splitting
Model Selection
Model Training
Model Evaluation
Hyperparameter Tuning
Prediction
Deployment
Monitoring & Retraining
Data Normalization
Data normalization is a preprocessing technique used in machine learning to scale numerical features to
a common range, so that no feature dominates the model due to its large magnitude.
Why Data Normalization is Needed
• Different features may have different units and scales
o Example: Age (1–100), Income (₹10,000–₹10,00,000)
• Improves model accuracy
• Speeds up model training
• Essential for distance-based and gradient-based algorithms
Common Normalization Techniques
1. Min–Max Normalization
Scales values to a fixed range, usually [0, 1]
Formula:
𝑋 − 𝑋min
𝑋′ =
𝑋max − 𝑋min
Example:
• Original value = 50
• Min = 0, Max = 100
• Normalized value = 0.5
Used in: Neural Networks, K-NN
2. Z-Score Normalization (Standardization)
Transforms data to have mean = 0 and standard deviation = 1
Formula:
𝑋−𝜇
𝑋′ =
𝜎
Used in: Linear Regression, SVM, PCA
3. Decimal Scaling
Moves decimal point based on maximum value
Formula:
𝑋
𝑋′ =
10𝑗
Used when: Data range is very large
4. Max Normalization
Divides each value by the maximum value
Formula:
𝑋
𝑋′ =
𝑋max
When to Use Normalization
• Algorithms based on distance:
o K-NN
o K-Means
• Algorithms using gradient descent:
o Logistic Regression
o Neural Networks
• Feature magnitudes vary significantly
Normalization vs Standardization
Aspect Normalization Standardization
Range Fixed (0–1) No fixed range
Aspect Normalization Standardization
Mean Not centered Mean = 0
Std Dev Not 1 Std Dev = 1
Sensitive to Outliers Yes Less
Example Method Min–Max Z-Score
1. Linear Regression
Linear Regression is a supervised learning algorithm used to predict a continuous value by establishing a
linear relationship between input variables and the output.
Example: Predicting house price based on area.
2. Decision Tree
A Decision Tree is a supervised learning algorithm that makes decisions using a tree-like structure of
nodes and branches, where each internal node represents a condition, and each leaf node gives the final
output.
Example: Loan approval decision.
3. K-Nearest Neighbors (K-NN)
K-NN is a supervised learning algorithm that classifies or predicts data points based on the majority class
or average of its K closest neighbors in the feature space.
Example: Classifying a student as pass/fail.
4. Support Vector Machine (SVM)
Support Vector Machine (SVM) is a supervised learning algorithm that finds an optimal hyperplane that
best separates data points of different classes with maximum margin.
Example: Spam vs non-spam email classification.
5. Neural Networks
Neural Networks are learning models inspired by the human brain, consisting of interconnected layers of
neurons that learn complex patterns through weighted connections and activation functions.
Example: Image and speech recognition.
Introduction to Decision Tree
A Decision Tree is a supervised machine learning algorithm used for classification and regression. It
represents decisions in the form of a tree-like structure, where each internal node tests a condition on
an input feature, branches represent the outcome of the test, and leaf nodes represent the final decision
or prediction.
Key Components of a Decision Tree
• Root Node: The topmost node representing the entire dataset.
• Internal Nodes: Nodes that perform a decision based on a feature.
• Branches: Outcomes of the decision rule.
• Leaf Nodes: Terminal nodes that give the final output (class or value).
How a Decision Tree Works
1. Select the best feature to split the data.
2. Divide the dataset based on feature values.
3. Repeat splitting recursively for each subset.
4. Stop when a stopping condition is met.
5. Assign a class label or value at the leaf node.
Splitting Criteria
• Entropy & Information Gain (ID3)
• Gini Index (CART)
• Chi-square (CHAID)
Advantages
• Easy to understand and interpret.
• Works with both numerical and categorical data.
• Requires little data preprocessing.
• Can model non-linear relationships.
Disadvantages
• Can overfit the data.
• Sensitive to small data changes.
• Less accurate compared to ensemble methods.
Applications
• Medical diagnosis
• Credit risk analysis
• Customer segmentation
• Fraud detection
Applications of Machine Learning
Machine Learning is widely used to analyze data, make predictions, and automate decisions in many
fields. Some important applications are:
1. Healthcare
• Disease Diagnosis: Predict diseases like cancer, diabetes, and heart conditions using patient
data.
• Medical Imaging: Detect abnormalities in X-rays, MRIs, or CT scans using ML algorithms.
• Drug Discovery: Accelerates development of new drugs by analyzing molecular structures and
clinical data.
• Personalized Treatment: Suggests personalized medication or therapy based on patient history.
2. Finance
• Fraud Detection: Detects unusual transactions in banking and credit cards.
• Stock Market Prediction: Predicts stock trends using historical data and market indicators.
• Credit Scoring: Evaluates creditworthiness of loan applicants.
• Algorithmic Trading: Automates buying and selling based on market analysis.
3. E-commerce and Retail
• Recommendation Systems: Suggests products on platforms like Amazon or Netflix based on user
behavior.
• Customer Segmentation: Groups customers for targeted marketing campaigns.
• Demand Forecasting: Predicts product demand to manage inventory efficiently.
4. Transportation
• Self-Driving Cars: ML algorithms process sensor and camera data to navigate vehicles.
• Traffic Prediction: Predicts traffic congestion and suggests optimal routes.
• Fleet Management: Optimizes routes and fuel consumption in logistics.
5. Natural Language Processing (NLP)
• Chatbots and Virtual Assistants: Siri, Alexa, and Google Assistant use ML to understand and
respond to human queries.
• Language Translation: Tools like Google Translate use ML for accurate translation.
• Sentiment Analysis: Analyzes social media posts or reviews to determine public opinion.
6. Image and Video Processing
• Face Recognition: Security systems use ML to recognize faces.
• Object Detection: Identifies objects in images for autonomous vehicles, surveillance, and
robotics.
• Video Analytics: Monitors public places or factories for safety and productivity.
7. Cybersecurity
• Intrusion Detection: Detects malicious activities in networks.
• Spam Filtering: Email services use ML to filter spam or phishing emails.
• Threat Prediction: Predicts potential cyber attacks using historical patterns.
8. Agriculture
• Crop Disease Detection: ML analyzes plant images to detect diseases early.
• Yield Prediction: Predicts crop output based on soil, weather, and other data.
• Precision Farming: Optimizes irrigation, fertilizer, and pesticide use.
9. Manufacturing
• Predictive Maintenance: Predicts machine failure to reduce downtime.
• Quality Control: Detects defects in products using computer vision.
• Supply Chain Optimization: Improves inventory and production planning.
10. Entertainment
• Content Recommendation: Suggests movies, music, or shows based on user preference.
• Game AI: Creates intelligent NPCs and adaptive gaming experiences.
Definition
• Artificial Intelligence (AI):
AI is the broader concept of machines being able to perform tasks that normally require human
intelligence. It includes reasoning, learning, problem-solving, perception, and language
understanding.
• Machine Learning (ML):
ML is a subset of AI that allows machines to learn from data and improve performance
automatically without being explicitly programmed for every task.