Comparison of Prediction Models for Diabetes: Accuracy Analysis of Logistic Regression, Random Forest, and Back Propagation Neural Network

Diabetes is one of the major non-communicable diseases and the second largest chronic disease, which has a huge impact on human health, medical resources, and social economy. In this study, the effectiveness of three machine learning models was verified, including Logistic Regression model (LR model), Random Forest model (RF model), and Back Propagation Neural Network model (BPNN model) in diabetes prediction, provided important reference and guidance for the selection of diabetes prediction models, and is expected to provide more reliable support for clinical decision-making. In the model evaluation, this study used the diabetes data set (containing 1879 samples and 46 feature variables). Firstly, 10 core variables were screened out by calculating the Spearman coefficient. Secondly, taking accuracy as the main indicator, the LR, RF, and BPNN models were analyzed in detail. The results show that the RF model performed best in this study and achieved the highest accuracy. Therefore, in the field of diabetes prediction, the RF model may be more suitable for this dataset. However, this study has not systematically compared the performance of different models under multi-parameter configuration, and further exploration can be carried out through parameter optimization and model tuning to clarify the optimal model architecture, and on this basis, the prediction performance of the three models can be compared and analyzed, to construct a more scientific and reliable evaluation system. The study also suggests that the performance of the RF model can be further optimized.