A ROBUST MACHINE LEARNING APPROACH USING MOTE AND HYPERPARAMETER OPTIMIZATION FOR DIABETES PREDICTION USING WEARABLE IOT DEVICE DATA

The introduction of wearable IoT devices to the market has accelerated the speed of continuous health monitoring and early diagnosis of chronic conditions such as diabetes. This paper proposes a strong machine learning approach to diagnosing diabetes using wearable IoT devices’ data collected. The suggested method integrates Synthetic Minority Oversampling Technique (SMOTE) for minority class and features that intrinsically come with medical datasets, and employs hyperparameter optimization to elevate model performance. By using data preprocessing skills and feature engineering methods, the study is guaranteed to have the best model input representation. Ensemble learning techniques such as XGBoost and LightGBM are utilized for building robust predictive models that are further improved by a stacking ensemble. Hyperparameter tuning is done with Randomized SearchCV to perfect the models for maximum accuracy and efficiency in the medical datasets. Experimental results of the IoT real-world dataset are carried on to the proposal sustaining an accuracy of 88% with other significant precision and recall values in both diabetic and non-diabetic classes. The research pinpoints the ability of wearable IoT devices to work together with advanced machine learning algorithms to change diabetes management which can bring about faster treatments and reduced patient income. The findings talk about the possibility of creating real-time data analytics systems supported by IoT technology for clinical decision-making processes.