Performance Assessment of Neural Networks in Medical Treatment Cost Estimation

Abstract

This study explores the use of simulations to analyze the performance of predictive neural network models applied in healthcare, specifically in estimating medical treatment costs. The simulation uses synthetic data generated for 30 days, with variables such as treatment cost, disease severity, and patient waiting time. Three neural network models — Feedforward NN, LSTM NN, and 1D CNN — are employed for cost prediction, and their performance is evaluated using the Mean Absolute Error (MAE), Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) metrics. In addition to numerical evaluation, a series of detailed visualizations are created to analyze the relationships between variables and to compare the accuracy of each model. The results indicate that the 1D CNN achieves the best performance, demonstrating its potential for real-life clinical applications. The study also highlights the utility of simulations for testing predictive methodologies in fields where real-world data may be limited.