mohammadsammour

Detects early-stage bearing failures using unsupervised anomaly detection. Extracts vibration features (RMS, Std, Peak) from sensor data, models healthy behavior with a Gaussian distribution, and flags anomalies via log-probability scoring — providing hours of warning before actual failure.

A machine learning pipeline for binary classification of gallstone disease using clinical and metabolic features. Trained and evaluated four models with hyperparameter tuning, MLflow experiment tracking, and explainability via SHAP and LIME.

From-scratch K-Means clustering implementation in Python using only NumPy. Covers centroid assignment, mean recomputation, and random initialization. Applied to real-world image compression, reducing a 24-bit RGB image to 16 colors and achieving a ~6× reduction in storage size.

Deep-learning waste classifier for separating organic and recyclable items. It compares 12 transfer-learning experiments across ResNet18, VGG-16, EfficientNet-B0, and MobileNetV2 using accuracy, precision, recall, F1-score, model size, training time, and overfitting analysis, then serves the best model through Streamlit for practical image testing.