Hybrid Deep Learning Model for Diabetes Prediction and Automated Diet Recommendation Using Ensemble Classifiers and Optimized Algorithms

Abstract

Nowadays, Non-Insulin-Dependent Diabetes Mellitus (NIDDM) is one of the most life-threatening and multifaceted illnesses. Many individuals are afflicted with numerous ailments because of insufficient nutrition for diabetic patients. Numerous machine learning practices and approaches adopted and suggested by researchers for predicting diabetes in the current past. In this paper, various classifiers including ensemble voting classifier, artificial neural network, AdaBoost, Random Forest, Support Vector Machine, Gradient Boost and deep learning models like Tabnet, XgBoost, is used to detect diabetic patients from reported dataset and recommend the required calories for a diabetic patient in the form of diet. Further, the results were equated with the prevailing models in terms of classification and recommendation. The PIMA India diabetes dataset is considered to check the performance of the model, which reveal that the Voting Classifier achieved a remarkable accuracy rate of 98.85%. The optimized hybrid deep learning model is designed to perform the recommendation process that helps to select the meals in breakfast, lunch and dinner for diabetes patients. Various performance metrics such as recall, F1 score, accuracy and precision are analyzed on the proposed method. This paper proposed a novel automated Diet Recommendation to recommend the nutrition diet plan for diabetic patients depending on patient’s health and medical records. The diet-recommendation system is examined using a hybrid deep learning model called Triple Attention-based Gated Stacked LSTM (TriAtt-St-LSTM). Conclusion: The proposed work has achieved a highest accuracy of 98.09% for the UCI dataset and 99.45% for the diet recommendation system dataset.