imblanced-data

🧠 9 real-world ML projects from AI/ML internship at Stemeta.ai | Classification · NLP · Fraud Detection · Flask API | Python · Scikit-Learn

End-to-end binary classifier on 150K+ records. Used SMOTE to resolve severe class imbalance (8K → 111K samples), benchmarked 5 models, achieved 89.1% accuracy and AUC 0.818 with Random Forest. Production-ready with pickle deployment.

This repository contains a machine learning pipeline for predicting bank marketing campaign success using the Bank Marketing Dataset. It includes data preprocessing, model training (Logistic Regression and Random Forest), and threshold optimization to improve recall for the minority class. The final model is evaluated using precision, recall.

Fraud detection case study tackling extreme class imbalance using Generative AI (Gemini) to create synthetic minority-class samples.

ICU patient mortality prediction using machine learning. The project explores Logistic Regression, KNN, and Random Forest models with focus on handling class imbalance, optimizing recall for mortality detection, and applying threshold tuning for clinical decision support.

Rare-object detection of white dwarfs in Gaia DR3. Uses spectroscopically confirmed labels (Gaia-SDSS) to avoid label leakage from photometric catalogues, field-star negatives from an ADQL pull, and logistic regression, random forest and XGBoost evaluated on PR-AUC.

Fraud detection data preprocessing pipeline with SMOTE balancing, PCA dimensionality reduction, and feature engineering.

Predicting diabetes risk from CDC health indicators: Logistic Regression, Random Forest, LightGBM and MLP compared with recall-focused threshold tuning and SHAP explainability. Machine Learning I group project, Hanyang University.

An end-to-end Machine Learning project to predict machine failures and support proactive maintenance using sensor data.