This repository contains a comprehensive suite of empirical studies and implementations of various Machine Learning algorithms. The primary focus of this project is the end-to-end analytical pipeline: from raw data preprocessing and feature engineering to model evaluation and hyperparameter tuning across diverse, real-world datasets (medical diagnostics, customer segmentation, etc.).
The project investigates both supervised and unsupervised learning paradigms, providing a comparative analysis of computational efficiency and predictive accuracy:
- Supervised Learning (Classification & Regression):
- Decision Trees: Applied to survival analysis (Titanic dataset) and demographic performance prediction. Focus on entropy/Gini impurity optimization and tree pruning techniques.
- K-Nearest Neighbors (K-NN): Implemented for medical diagnostics (Parkinson's disease detection) and chemical analysis (Wine quality). Includes distance metric evaluation and dimensionality considerations.
- Regression Models: Linear and Logistic Regression for continuous prediction and binary classification (Diabetes progression), including feature scaling and multicollinearity analysis.
- Unsupervised Learning (Clustering):
- K-Means: Extensive empirical evaluation across multiple behavioral and biological datasets. Focus on centroid initialization, convergence analysis, and optimal cluster selection (Elbow method).
- Language: Python 3
- Libraries:
scikit-learn,pandas,NumPy,Matplotlib,Seaborn - Methodologies: Cross-validation, Feature Engineering, Exploratory Data Analysis (EDA), Performance Metrics Evaluation (Accuracy, Precision, Recall, F1-Score).
This suite demonstrates a pragmatic, engineering-focused approach to data science. By applying different algorithmic architectures to the same or varying data domains, it highlights the trade-offs between model interpretability, training complexity, and generalization capabilities.