Deep Learning-Based Early Diabetes Risk Prediction Using Survey Data: A Hyperparameter Optimization Approach

Radwa Ahmed Osman

Abstract: Diabetes must be detected accurately and early to ensure successful prevention and control. This article describes a deep learning-based technique that use a Deep Convolutional Neural Network (DCNN) to categorize diabetes risk using health indicators from the 2015 Behavioral Risk Factor Surveillance System. Three versions of the dataset were tested: a multiclass dataset with three diabetes states (no diabetes, prediabetes, and diabetes), a binary classification version, and a balanced binary version with an equal proportion of diabetic and non-diabetic patients. The suggested DCNN model was trained on 21 health-related survey characteristics, such as BMI, physical activity, smoking status, and overall health perception. Normalization and class balancing were performed during preprocessing. An intensive hyperparameter tuning procedure was carried out to guarantee that the model obtained the lowest loss and highest classification accuracy. This stage was crucial since the choice of suitable hyperparameters-such as learning rate, batch size, number of filters, kernel size, and number of epochs-had a direct impact on the model's capacity to learn significant patterns from data while avoiding underfitting or overfitting. The improved DCNN outperforms traditional machine learning classifiers in terms of accuracy, recall, and F1-score across all dataset versions. Furthermore, feature importance analysis revealed the most significant risk variables involved in diabetes prediction. These findings demonstrate that carefully tuning hyperparameters in deep learning models can significantly enhance predictive performance, thereby supporting early detection efforts and informing public health interventions.

Keywords: Diabetes Prediction, optimization, Deep learning, 1D Convolutional Neural Network, Health Indicator, Lifestyle and Clinical Indicators, multi-class classification

How to Cite?: Radwa Ahmed Osman, "Deep Learning-Based Early Diabetes Risk Prediction Using Survey Data: A Hyperparameter Optimization Approach", Volume 11 Issue 11, November 2022, International Journal of Science and Research (IJSR), Pages: 1549-1558, https://www.ijsr.net/getabstract.php?paperid=SR221111091934, DOI: https://dx.doi.org/10.21275/SR221111091934