Glucose level prediction is an important area of research, but the application of machine learning models often faces challenges such as overfitting and limited data. This study aims to evaluate the performance of seven regression models—Linear Regression, Lasso, SVR, Decision Tree, Random Forest, Gradient Boosting, and K-Nearest Neighbors in predicting glucose levels on a small dataset that has undergone preprocessing and dimensionality reduction using Principal Component Analysis (PCA). The evaluation was conducted using the R-squared, Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) metrics. The results showed that most models suffered from severe overfitting, where good performance on the training data was not followed by performance on the test data, as reflected by negative R-squared values. Although Random Forest produced relatively lower errors (MSE = 0.009; RMSE = 0.095), the overall performance remained inadequate due to data limitations. These findings confirm that small sample sizes and overfitting simultaneously hamper the generalizability of models and emphasize the need for more robust approaches in biomedical applications with limited data.
Copyrights © 2025