threshold-optimization

Cutoffs for max. multiclass F1-score, etc.

🛰️ Production-ready ML system for geomagnetic storm prediction | 98% AUC, 70% recall | Threshold-optimized ensemble with real-time inference | 29-year dataset (1996-2025) | NOAA SWPC operational standards | Complete MLOps pipeline

An end-to-end cost-sensitive customer churn prediction system that combines machine learning, business cost optimization, threshold tuning, and an interactive Streamlit dashboard to prioritize customer retention and reduce potential revenue loss.

Cost-sensitive loan default prediction using Python and machine learning, with threshold optimization, business cost simulation, model interpretation, and responsible AI considerations.

Rescuing Moore’s Law via Room-Temperature Soliton Integration in Standard Silicon.

End-to-end credit card fraud detection using Random Forest and LightGBM, with class imbalance handling, threshold optimization, and SHAP-based model interpretability.

End-to-end fraud detection pipeline on 500K+ IEEE-CIS transactions, combining imbalanced ML, cost-sensitive threshold tuning, SQLite storage, real-time scoring, and live risk monitoring.

B2B sales lead quality prediction using XGBoost classifier. Achieves 81.06% ROC AUC and 84.74% recall on 7,420 IT sales leads. Handles class imbalance, high-cardinality categoricals, and missing data through frequency encoding and threshold optimization. Includes statistical analysis, cross-validation, feature importance, and business insights.

Build a Gardner Logic-Bridge IP core for threshold-programmable, room-temperature soliton logic in standard silicon

Credit risk modeling with HistGradientBoosting, featuring evaluation, SHAP explainability, threshold optimization, and high-risk client analysis.

End-to-end supervised ML project predicting Indian cricket team match outcomes using historical data. Covers EDA, feature engineering, Logistic Regression, KNN, Naive Bayes, and Decision Tree (with GridSearchCV tuning) — with actionable BCCI strategy recommendations.

Exploratory financial fraud detection ML pipeline built to study class imbalance, feature behavior, and threshold tradeoffs. Uses XGBoost on a large transaction dataset to analyze recall-precision dynamics, data shortcuts, and system limitations. Educational, not production-ready.

Built a machine learning model to predict telecom customer churn using classification techniques and SHAP explainability. Optimized performance through tuning and translated results into actionable customer retention insights.

Home Credit Risk Prediction (AUC 0.794 / Private LB Top 15%)

Churn prediction optimized for business value: cost matrix, threshold optimization, and sensitivity analysis.

Visualize binary classifier performance with operating profile plots: score histograms + TPR/FPR/accuracy metrics across all decision thresholds. Python tool for model validation, threshold tuning, ROC analysis, calibration audits

Built and deployed an employee attrition prediction application using Scikit-learn and Streamlit. Engineered HR-specific features, applied feature normalization, benchmarked multiple ML algorithms, optimized model hyperparameters and decision thresholds, and integrated the final Logistic Regression pipeline for real-time attrition risk prediction.

Which patients get a confirmatory test under limited clinic capacity: a cost-derived threshold cuts cost by ~97%

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.

End-to-end credit card fraud detection using leakage-safe ML pipelines, imbalance handling, model comparison, failure analysis, and cost-sensitive threshold optimization.