/CSE-472

Machine Learning Sessional

Primary LanguageJupyter Notebook

CSE 472: Machine Learning Sessional

This repository contains the assignments and solutions for the CSE 472: Machine Learning Sessional course. Each offline session focuses on key machine learning concepts and algorithms, with hands-on implementations and experimentation.

Contents

  1. Offline 1: Data Preprocessing and Feature Engineering
  2. Offline 2: Ensemble Learning with Logistic Regression
  3. Offline 3: Neural Network and Backpropagation
  4. Offline 4: PCA and Expectation-Maximization

Offline 1: Data Preprocessing and Feature Engineering

File: 1905077.ipynb

Introduction

This assignment covers data preprocessing and feature engineering for machine learning models. It includes tasks like cleaning raw data, handling missing values, normalizing datasets, and feature selection.

Key Tasks

  • Import and preprocess the "IBM HR Analytics Employee Attrition & Performance" dataset.
  • Handle missing values, redundancy, and data normalization.
  • Convert categorical variables into numerical representations.
  • Perform correlation analysis to identify important features.
  • Prepare the dataset for a machine learning pipeline.

Dataset

Offline 2: Ensemble Learning with Logistic Regression

File: 1905077.ipynb (Datasets in the data/ folder)

Introduction

This assignment focuses on implementing Logistic Regression (LR) from sratch, ensemble learning techniques using bagging and stacking and LR the base classifier.

Key Tasks

  • Preprocess datasets to standardize input formats.
  • Implement Logistic Regression (LR) as the base learner.
  • Implement Bagging with 9 LR models and Stacking with LR as the meta-classifier.
  • Create a simple majority voting-based ensemble for comparison.
  • Evaluate model performance using metrics and violin plots.

Datasets

  1. Telco Customer Churn Dataset
  2. UCI Adult Dataset
  3. Credit Card Fraud Detection Dataset

Offline 3: Neural Network and Backpropagation

File: 1905077.ipynb

Introduction

This assignment involves implementing a Feed-Forward Neural Network (FNN) from scratch for apparel classification.

Key Components

  • Dense Layer: Fully connected layer.
  • Batch Normalization: Normalizes the input for each layer.
  • ReLU Activation: Activation function for hidden layers.
  • Dropout: Regularization to prevent overfitting.
  • Adam Optimizer: Adaptive moment estimation for weight updates.
  • Softmax Regression: For multi-class classification.

Key Tasks

  • Modularize the implementation to allow flexibility in architecture.
  • Implement backpropagation and mini-batch gradient descent for training.
  • Train and evaluate the FNN using the provided dataset.

Offline 4: PCA and Expectation-Maximization

Files: em.ipynb and pca.ipynb

Part 1: Principal Component Analysis (PCA)

Dataset: pca_data.txt

  • Contains 1000 rows and 500 columns representing 1000 sample points with 500 features each.

Key Tasks

  1. Perform PCA for dimensionality reduction.
  2. Project data along the two eigenvectors corresponding to the highest eigenvalues.
  3. Create a 2D scatter plot for visualization.
  4. Generate UMAP and t-SNE plots using library functions for comparison.

Part 2: Expectation-Maximization (EM)

Dataset: em_data.txt

  • Represents the number of children in 1000 families, with some given family planning advice.

Key Tasks

  1. Implement the EM algorithm for Poisson mixture models.
  2. Estimate:
    • Mean number of children in families with and without family planning.
    • Proportion of families with and without family planning.

How to Use

Clone the Repository

git clone <repository_url>
cd <repository_directory>

### Run the relevant .ipynb files under the folders