A Robust Disease Prediction System Using Hybrid Deep Neural Networks

Abstract

Healthcare data analysis is one of the attractive research topics in the research community. Researchers pay more attention to developing an efficient and accurate healthcare system. Numerous techniques are evolved in the past decade and research is still in progress to enhance the performance of healthcare systems. The initial stage of healthcare data analysis systems employs statistical methods to predict or detect diseases from healthcare data. However, due to the data heterogeneity, and large volume the statistical methods produces incorrect results which affect the healthcare system performance. The technological development and penetration of artificial intelligence in the healthcare domain presented various approaches and solutions for traditional issues. Early disease prediction is crucial for improving patient outcomes and reducing healthcare costs. This study explores the efficacy of various hybrid deep learning techniques in predicting the onset of chronic diseases, focusing on diabetes, cardiovascular diseases, and certain types of cancer. We compared the performance of several hybrid algorithms, including Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), Gradient Boosting-Neural Network (GB-NN), and Autoencoder-Support Vector Machine (AE-SVM), using a diverse dataset of 100,000 patient records. Our results indicate that the CNN-LSTM model achieved the highest accuracy (95.3%) in predicting disease onset within a 5-year window, followed closely by GB-NN (94.1%) and AE-SVM (92.8%). The study also identified key predictive factors and discussed the potential integration of these hybrid models into clinical decision support systems. Our findings suggest that hybrid deep learning approaches can significantly enhance early disease detection and intervention strategies in healthcare settings, outperforming traditional single-algorithm methods.