Development of a Neural Network-Based Predictive System for Automated Diabetes Detection Using Clinical Data
Udhayakumar S1, Dr Kamalakannan S2
1Assistannt Professor, Dept. of ECE, KGiSL Institute of Technology Coimbatore, TN, India
udhayame10@gmail.com
2Associate Professor, Dept. of ECE, KGiSL Institute of Technology Coimbatore, TN, India
kamalsphd@gmail.com
Abstract— Diabetes is a widespread chronic disease that requires timely diagnosis and management to reduce severe health complications like cardiovascular issues, kidney failure, and neuropathy. Traditional diagnostic methods rely heavily on invasive procedures and expert analysis, which may not be readily accessible in resource-limited settings. This paper explores the design of a neural network-based predictive system that automates diabetes detection using patient clinical data. By employing machine learning techniques, specifically feedforward neural networks, the model effectively learns patterns in features such as glucose levels, BMI, insulin, age, and family history.
The system incorporates advanced preprocessing methods, including feature normalization and outlier removal, to enhance model accuracy. Hyperparameter tuning and regularization techniques, like dropout, prevent overfitting and ensure generalizability. Experimental results show the model achieves significant predictive accuracy compared to traditional methods. The application of neural networks demonstrates the capability to provide non-invasive, cost-effective, and efficient diabetes prediction tools. Future enhancements, including integration with IoT devices and explainable AI techniques, promise even greater impact by enabling real-time diagnosis and fostering trust among healthcare providers and patients.
Keywords—Feedforward neural networks, feature normalization, Hyperparameter Tuning, Non-Invasive Diagnostic Tools.
