Insurance Claim Prediction using Deep Learning for Oromia Insurance S.C

Abstract

This study outlines the development and implementation of a deep learning model for prediction insurance claims. This study, we present a deep learning models for the task of insurance claim prediction. We compare the performance of multiple architectures, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Networks (CNN), and their hybrid combinations (LSTM-CNN, GRU-CNN, FFNN-CNN), alongside a Fully CNN model. Our findings demonstrate that hybrid models consistently outperform standalone architectures. Specifically, the LSTM-CNN model, optimized with the Adam optimizer at a learning rate of 0.001, achieved superior results across all evaluation metrics. It attained the lowest validation loss (0.0004), Mean Squared Error (0.0002), Mean Absolute Error (0.0041), Root Mean Squared Error (0.0142), and Mean Absolute Percentage Error (4.9097%), alongside the highest R² score (0.9977), accuracy (99.18%), and precision (99.18%). The results strongly suggest that hybrid deep learning models, particularly LSTM-CNN, coupled with the Adam optimizer, offer a highly effective and robust framework for advancing the accuracy of insurance claim predictions, thereby supporting well administrative for insurers. This study addresses the gap in Ethiopian insurance claim prediction by applying hybrid deep learning models to Oromia Insurance datasets