Featured Projects

Based on a comprehensive analysis of the CIC-IDS-2017 dataset, a machine learning pipeline was developed to detect and analyze network intrusions. The project successfully produced a highly accurate

XGBoost model for attack detection, achieving 99.87% accuracy and a 99.61% F1-Score.

Model interpretability techniques identified key predictive features, including

`Destination Port` and `Init_Win_bytes_forward`. Furthermore, a detailed forensic investigation of the

DoS Hulk attack identified its unique signature, which includes an abnormally high `Max Packet Length` (10.53x higher than benign traffic) and `Flow Duration` (5.34x higher). These findings provide a strong, data-driven basis for real-time intrusion detection and incident response.

This project implements a deep learning solution for classifying handwritten digits (0-9) from the MNIST dataset using PyTorch. The implementation features two convolutional neural network (CNN) architectures designed to achieve high accuracy on digit recognition tasks.

Key Features:
Dual CNN Architectures: Basic ImprovedNN and enhanced EnhancedCNN models
Comprehensive Training Pipeline: Complete data preprocessing, training loops, and validation
Performance Optimization: Learning rate scheduling, dropout regularization, and GPU acceleration
Visualization Tools: Training metrics plots and sample image display functions
Model Persistence: Save and load trained model weights for future use
Results:
Achieves >99% accuracy on MNIST test set
Implements modern deep learning best practices
Includes detailed documentation and code explanations

This project predicts hourly bike rental demand using AutoGluon, an automated machine learning framework. Developed as part of the AWS Machine Learning Scholarship Program, it uses weather, time, and seasonal data to forecast bike usage. The project includes feature engineering, hyperparameter tuning, and Kaggle-formatted submissions, showcasing how AutoML can solve real-world transportation problems efficiently.