Medical Insurance Cost Prediction Using Gradient Boosting Regression: A Machine Learning Approach

Abstract

This paper highlights the point that correct forecasting of the expense of medical insurance is essential in the better decision-making of individuals, insurers, and policymakers to efficiently allocate resources in the dynamically changing environment of healthcare financing. While recent studies have extensively explored machine learning (ML) approaches for medical insurance cost prediction, there remains a critical need to improve their accuracy and reliability, driving the pursuit of more effective methods to enhance the precis. In the context of these caveats, there exists a research gap to which this investigation attempts to contribute by proffering an ML method using the Gradient Boosting Regressor (GBR), through which one can enhance the level and quality of prediction for medical insurance expenses. To deal with this, this study presents a GBR base approach for predicting medical insurance costs from a dataset of 1,339 samples with seven features, such as age, sex, BMI, smoking, and region. The dataset from Kaggle offers thorough coverage of the factors affecting medical insurance costs. Our approach involves extensive preprocessing of the data, including one-hot encoding for categorical features, followed by training, validation, and evaluation of the model using an 80/20 train-test split. We rigorously evaluated the performance of the GBR model using the metrics of Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared (R²), Mean Absolute Percentage Error (MAPE), and Explained Variance Score (EVS). Experimental outcome further establishes that the best-performing model is the GBR model, based on obtained results as reflecting better predictive accuracy. Comparison with Linear Regression, Random Forest, Support Vector Regression, K-Nearest Neighbors, and Neural Networks further established that the best precision (0.908), recall (0.903), and F1-score (0.899) is achieved by the GBR model. These findings support the effectiveness of the GBR model as a powerful tool for capturing nonlinear patterns of relationship underlying the data, for predicting medical insurance costs. This research highlights the usefulness of sophisticated techniques of machine learning for improving predictive modeling of healthcare finances.