FinTech Reconciliation Break Classifier (Pair-Level)

This repository contains a trained model that predicts whether a bank↔broker reconciliation pair is a TRUE_BREAK (1) or FALSE_BREAK (0).

Features Used

['large_amt_mismatch', 'amt_diff', 'amt_pct', 'amt_diff_abs', 'amt_ratio', 'settlement_gap_days', 'is_weekend', 'is_settlement_weekend']

Best Run (selected by PR-AUC)

  • Experiment: exp_001_baseline
  • PR-AUC: 1.000
  • ROC-AUC: 1.000
  • F1: 0.998

Files

  • model.joblib : trained sklearn pipeline
  • features.json: selected feature list
  • config.json : training metadata + best metrics
  • predict.py : minimal inference helper
Downloads last month
6
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support