Medical Claims Fraud Detection Dashboard

Overview

This dashboard presents a brief analysis of a medical claims fraud detection project. The analysis includes:

Supervised classification using Random Forest
Unsupervised anomaly detection using Isolation Forest
Clustering-based outlier detection using DBSCAN

Data was simulated to mimic a real-world scenario with 1,000 samples, 10 features, and a 5% fraud rate.

Class Distribution

Random Forest Analysis

The Random Forest model was trained to classify fraudulent claims using balanced class weights. Key evaluation metrics include the confusion matrix and ROC curve.

Confusion Matrix

ROC Curve

Isolation Forest Analysis

The Isolation Forest method detects anomalies without using labels during training. It flags potential fraudulent claims based on unusual patterns in the data.

Isolation Forest Report

              precision    recall  f1-score   support

           0       0.95      0.96      0.95       284
           1       0.08      0.06      0.07        16

    accuracy                           0.91       300
   macro avg       0.52      0.51      0.51       300
weighted avg       0.90      0.91      0.91       300

DBSCAN Clustering Analysis

DBSCAN clusters similar claims together while identifying noise points, which can correspond to suspicious or fraudulent claims.

DBSCAN Clustering

DBSCAN Report

              precision    recall  f1-score   support

           0       0.00      0.00      0.00       946
           1       0.05      1.00      0.10        54

    accuracy                           0.05      1000
   macro avg       0.03      0.50      0.05      1000
weighted avg       0.00      0.05      0.01      1000