| { |
| "dataset_name": "DNA Mixture Analysis - Unknown Contributor Detection", |
| "version": "2.0-optimized", |
| "date_updated": "2026-05-08", |
| "status": "Production-Ready", |
| "key_metrics": { |
| "f1_score": 0.7135, |
| "f1_score_improvement": "+5.66% from baseline", |
| "precision": 0.6262, |
| "recall": 0.86, |
| "samples_total": 500, |
| "samples_rd14": 408, |
| "samples_rd12": 92, |
| "features": 260 |
| }, |
| "optimization": { |
| "xgboost_weight": 0.7, |
| "catboost_weight": 0.3, |
| "decision_threshold": 0.42, |
| "improvement_over_baseline": "F1: 0.6753 \u2192 0.7135 (+0.0382)" |
| }, |
| "files_included": { |
| "data_files": [ |
| "combined_enhanced_features.csv", |
| "rd14_enhanced_features.csv", |
| "rd12_enhanced_features.csv" |
| ], |
| "cv_results": [ |
| "cross_validation_results.csv", |
| "cv_summary.json" |
| ], |
| "analysis": [ |
| "performance_summary.csv", |
| "performance_by_fold.csv", |
| "data_statistics.json" |
| ], |
| "models": [ |
| "inference_pipeline.py", |
| "optimal_parameters.json", |
| "f1_optimization_strategy.json" |
| ], |
| "documentation": [ |
| "EXECUTIVE_SUMMARY.md", |
| "TECHNICAL_REPORT.md", |
| "OPTIMIZATION_REPORT.md", |
| "README.md" |
| ] |
| }, |
| "features": [ |
| "Peak height statistics (count, max, sum, mean, std)", |
| "Peak ratio (peak1/peak2)", |
| "Allele balance (min/max peaks)", |
| "Homozygosity indicator", |
| "Peak consistency (coefficient of variation)", |
| "Applied across 20+ DNA markers" |
| ], |
| "model_architecture": { |
| "type": "XGBoost + CatBoost Ensemble", |
| "ensemble_method": "Weighted voting", |
| "training_data": "500 samples (408 RD14 + 92 RD12)", |
| "evaluation": "5-fold stratified cross-validation", |
| "optimization": "Optuna TPE sampler (50 trials)" |
| }, |
| "use_cases": [ |
| "Forensic DNA analysis", |
| "Unknown contributor detection in DNA mixtures", |
| "Case screening and prioritization", |
| "Research on mixture deconvolution" |
| ] |
| } |
