A Comparative Performance Benchmarking of Deep Learning Architectures (LSTM-RNNs and Transformers) Versus Classical Machine Learning Approaches for Multi-Feature Diabetes Onset Prediction

• Onucha Emeka

  Author

Diabetes mellitus remains a prevalent global health challenge, with early and accurate onset prediction being critical for effective intervention and management. While machine learning (ML) approaches have demonstrated utility in diabetes prediction, comparative evaluations of classical ML against advanced deep learning architectures for multi-feature prediction remain limited. This study presents a comprehensive comparative benchmarking of deep learning architectures—specifically Long Short-Term Memory Recurrent Neural Networks (LSTM-RNNs) and Transformers—against classical ML approaches including XGBoost, Random Forest, and Logistic Regression for multi-feature diabetes onset prediction. Using the PIMA Indian Diabetes dataset, we implemented rigorous preprocessing, feature engineering, and hyperparameter optimization across all models. Performance evaluation employed multiple metrics including accuracy, precision, recall, F1-score, and AUC-ROC. Results demonstrate that the Transformer model achieved the highest predictive accuracy at 89.4%, outperforming LSTM-RNN (86.2%), XGBoost (84.1%), and Random Forest (82.3%). Feature importance analysis identified glucose levels, BMI, age, and insulin as the most influential predictors. The Transformer architecture exhibited superior capability in capturing complex, non-linear temporal dependencies within the multi-feature dataset. These findings provide practical guidance for selecting appropriate predictive models based on available computational resources and clinical requirements. This benchmarking framework offers a replicable methodology for healthcare analytics applications where early diabetes risk identification is paramount.

• PDF

Copyright (c) 2026 Onucha Emeka (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.