This repository showcases a series of machine learning projects that build core skills in statistical modeling, iterative optimization, and predictive analytics.
Across the projects, models are implemented from scratch to emphasize mastery of the mathematical and computational foundations of AI and machine learning, including gradient-based optimization, regularization, feature selection, and probabilistic modeling.
The progression of projects reflects a transition from fundamental regression methods to advanced regularized models and multiclass classification, demonstrating both technical depth and applied problem-solving.
- Built a linear regression model trained via mini-batch gradient descent to predict sales from advertising spend.
- Explored convergence dynamics, including coefficient trajectories and cost function decay.
Skills: Linear Regression • Gradient-Based Optimization • Vectorized Computation • MSE Evaluation
- Implemented L2-regularized regression to predict credit card balances from demographic and financial features.
- Conducted hyperparameter tuning using five-fold cross-validation to balance bias and variance.
Skills: Ridge Regression • Regularization • Batch Gradient Descent • Cross-Validation • Hyperparameter Tuning
- Combined lasso (L1) and ridge (L2) penalties for feature selection and coefficient stabilization.
- Performed a grid search over (λ, α) with cross-validation and visualized coefficient paths to study sparsity vs. stability trade-offs.
Skills: Elastic Net • Coordinate Descent • Feature Selection • Regularization • Model Selection
- Developed a ridge-penalized multinomial logistic regression model to classify individuals into five continental ancestries using principal component genetic features.
- Produced probability-based predictions and interpreted outputs for admixed populations, connecting model results to population genetics insights.
Skills: Multiclass Logistic Regression • Ridge Regularization • Batch Gradient Descent • Categorical Cross-Entropy • Probabilistic Modeling
-
Mathematical Foundations:
- Linear and logistic regression, regularization (L1, L2, Elastic Net)
- Gradient-based and coordinate descent optimization
- Cross-entropy and mean squared error loss functions
-
Practical Machine Learning:
- Feature preprocessing (centering, scaling, standardization)
- Model selection with cross-validation and hyperparameter tuning
- Feature selection and coefficient path analysis
-
Applied Data Science and AI Foundations:
- From classical regression to probabilistic classification
- Interpreting model behavior and connecting outputs to real-world phenomena
- Producing reproducible, from-scratch implementations that reinforce theoretical understanding
This repository demonstrates the ability to bridge theory and practice in machine learning by building, tuning, and interpreting models without reliance on high-level ML libraries.
It serves as a portfolio of applied AI foundations, illustrating readiness for data science, analytics, and model development roles.