CSE375 Machine Learning Course Lab Repository: A Complete Learning Path from Basic Algorithms to Deep Learning

This repository is the complete lab record for East West University's CSE375 Machine Learning course, maintained by Yeamin Rahman Mahid. It includes 11 progressive lab modules covering the full tech stack from traditional machine learning to modern deep learning. Each lab provides code implementations, visualization results, and performance evaluations, forming a structured and reproducible learning path that is of great reference value for systematically mastering machine learning practices.

• Original Author/Maintainer: Yeamin Rahman Mahid
• Affiliated Institution: Department of Computer Science and Engineering, East West University
• Source Platform: GitHub
• Original Title: CSE438-Machine-Learning
• Original Link: https://github.com/mahidrahman375/CSE438-Machine-Learning
• Release Date: June 8, 2026

The project uses a combination of mainstream tools from the Python ecosystem:

• Data Processing: NumPy, Pandas (data cleaning, feature engineering)
• Machine Learning Framework: Scikit-learn (traditional algorithm implementation)
• Deep Learning: TensorFlow/Keras (neural network construction)
• Visualization: Matplotlib, Seaborn (analysis chart generation)
• Development Environment: Jupyter Notebook, Google Colab (interactive development)

The 11 lab modules are divided into five stages:

• Basic Supervised Learning (1-2): Data preprocessing (missing value/outlier handling, encoding, scaling), implementation and hyperparameter tuning of k-NN, weighted k-NN, decision tree algorithms, analysis of evaluation metrics (accuracy, precision, recall, F1 score).
• Ensemble Methods and Model Evaluation (3-4): Ensemble strategies like Bagging, Random Forest, AdaBoost, XGBoost; comparison of K-fold/stratified K-fold cross-validation.
• Unsupervised Learning and Anomaly Detection (5-7): K-Means, hierarchical clustering, DBSCAN clustering (WCSS/silhouette coefficient evaluation), Apriori association rules, PCA/t-SNE dimensionality reduction, Z-score, Isolation Forest anomaly detection.
• Deep Learning Basics (8-9): Analysis of loss functions (mean squared error, cross-entropy, etc.) and activation functions (Sigmoid, ReLU, etc.) characteristics; comparison of convergence behaviors of optimizers like gradient descent and Adam.
• Neural Network Architecture Practice (10-11): Complete implementation of ANN/CNN (pooling, Dropout, regularization); research on performance and efficiency trade-offs of CNN variants (depthwise separable convolution, dilated convolution, group convolution).

After completing the course, you will have:

1. End-to-end machine learning workflow construction ability
2. Ability to independently implement and compare multiple algorithms
3. Strict model evaluation and validation ability
4. Ability to apply high-dimensional data clustering and dimensionality reduction
5. Understanding of optimization algorithms and training dynamics
6. CNN architecture design and analysis ability

Future Expansion: Plan to add Kaggle competition solutions, research-based experiments, deep learning projects, real-world deployment models, and paper reproductions. License: MIT License, allowing free use and modification. Usage Suggestions: Practice in the order of the labs; try modifying parameters/replacing datasets to deepen understanding; apply traditional ML evaluation methods to deep learning experiments to integrate methodologies.