A Comparative Analysis and Multiclass Diabetes Prediction Using a Soft Voting Hybrid AI Model Based on Machine Learning and Neural Networks

Objective The goal of this study is to improve the early detection of diabetes, prediabetes, and non-diabetes to improve patient outcomes and help healthcare professionals make decisions. The study proposes a new AI model that uses advanced machine learning and neural networks to combine different methods for making accurate predictions about diabetes and related conditions. The study aims to find a better, more reliable way to help with diabetes diagnosis and management, which could save time and money in the healthcare system. RESEARCH DESIGN AND METHODS The study uses a combined machine learning approach that merges traditional methods like Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XG-Boost) with neural networks such as Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and Feedforward Neural Networks (FNN). We use the LMCH multiclass diabetes dataset from Mendeley Data to train and test our model. The study uses 5-fold and 10-fold cross-validation to test the soft voting-based hybrid model's performance, making sure the results are strong and trustworthy. We use several performance metrics, such as accuracy, precision, recall, F1 score, and ROC-AUC, to see how well the hybrid AI model can make predictions. RESULTS The proposed soft voting-based hybrid model demonstrated high performance across various metrics. For diabetes classification, the model achieved an accuracy of 0.986, precision of 0.971, recall of 0.892, F1 score of 0.926, and ROC-AUC of 0.994. For prediabetes, values were 0.996 (accuracy), 0.996 (precision), 0.996 (recall), 1.000 (F1 score), and 0.996 (ROC-AUC). For non-diabetes, results included an accuracy of 0.982, precision of 0.983, recall of 0.996, F1 score of 0.989, and ROC-AUC of 0.995 under 10-fold cross-validation. CONCLUSION This study demonstrates the predictive power of a hybrid AI model that blends machine learning and neural network algorithms for diabetes, prediabetes, and non-diabetes. The soft voting ensemble method makes predictions much more accurate and helps healthcare professionals make better decisions. The study suggests that implementing these hybrid models in real-life settings could better assist patients, reduce healthcare costs, and enhance the accuracy of diabetes management. Future studies might examine integrating various demographic datasets, applying models in real-time in clinical settings, and simplifying models to make predictions more scalable, intelligible, and customized for each patient's requirements.