An Effective Data Sampling Procedure for Imbalanced Data Learning on Health Insurance Fraud Detection

Fraud detection has received considerable attention from many academic research and industries worldwide due to its increasing popularity. Insurance datasets are enormous, with skewed distributions high dimensionality. Skewed class distribution volume considered significant problems while analyzing insurance datasets, as these issues increase the misclassification rates. Although sampling approaches, such random oversampling SMOTE can help balance data, they also computational complexity lead a deterioration of model's performance. So, more sophisticated techniques needed classes efficiently. This focuses on optimizing learner for fraud by applying Fused Resampling Cleaning Ensemble (FusedRCE) effective in health detection. We hypothesized that meticulous followed guided data cleaning would improve prediction performance learner's understanding minority fraudulent compared other techniques. The proposed model works three steps. As first step, PCA is applied extract necessary features reduce dimensions data. In second hybrid combination k-means clustering used resample imbalanced Oversampling introduces lots noise A thorough performed balanced remove noisy samples generated during using Tomek Link algorithm third step. clears boundary between majority makes precise freer noise. resultant dataset four different classification algorithms: Logistic Regression, Decision Tree Classifier, k-Nearest Neighbors, Neural Networks repeated 5-fold cross-validation. Compared classifiers, FusedRCE had highest average rate 98.9%. results were measured parameters F1 score, Precision, Recall AUC values. obtained show method significantly better than any approach predicting greater accuracy 3x speed training.