A collection of machine learning and data science projects covering data preprocessing, model development, evaluation, deep learning, and production-oriented ML workflows.
A three-part deep learning project exploring FCNN implementation from scratch, learned representations, spatial permutation, gradient flow, and training robustness.
- NumPy implementation from scratch
- PyTorch-based MNIST experiments
- Deep FCNN on Tiny ImageNet
- Vanishing-gradient analysis
- Training ablation study
More details: fully-connected-neural-networks/
A production-oriented fraud detection system built on the Kaggle Credit Card Fraud Detection dataset, covering exploratory analysis, imbalance handling, model comparison, threshold optimization, interpretability, experiment tracking, and API deployment.
- Exploratory data analysis and evaluation setup
- Time-based feature engineering and robust scaling
- Logistic Regression, LightGBM, and XGBoost
- Comparison of class weighting, SMOTE, and random undersampling
- Threshold optimization using business cost
- SHAP-based model interpretability
- MLflow experiment tracking
- FastAPI prediction API with Pydantic validation
- Automated tests
- Reusable training and prediction pipeline
More details: fraud-detection/