Optimizing Deep Learning using Feature Selection for Accurate Health Insurance Claim Classification

Abstract

Accurate prediction of health insurance claims is essential for improving hospital service quality, financial management, and operational decision-making. Health insurance claim data contains multiple attributes that may affect the performance and efficiency of classification models. Data mining techniques, particularly deep learning, are widely applied because of their ability to identify complex patterns in large datasets. However, the use of numerous attributes can increase model complexity and computational requirements. Therefore, feature selection can be applied to reduce data dimensionality while retaining relevant information for the classification process. This study aims to optimize a deep learning model using the Weight by User Specification method as a feature selection technique for classifying BPJS Health insurance claims. The dataset consists of 2,388 BPJS Health insurance claim records from Kumala Siwi Hospital, with variables including total rate, hospital rate, profit, diagnosis, and INACBG. The research process follows data preprocessing, data transformation, deep learning classification, feature selection, and model evaluation using a confusion matrix. Two classification models were evaluated: deep learning without feature selection and deep learning combined with Weight by User Specification. The results show that the deep learning model without feature selection achieved an accuracy of 98.87%, while the optimized model achieved 99.20%, representing an improvement of 0.33%. The classification error also decreased from 1.13% to 0.80%. These findings indicate that integrating feature selection with deep learning can improve classification accuracy and reduce classification error in BPJS Health insurance claim prediction. Therefore, weight by user specification combined with deep learning provides an effective approach for health insurance claim classification.