Deep Learning for Prediction of Cardiovascular Disease

Abstract

This study compares three deep learning algorithms for cardiovascular disease risk prediction. While RBFN boasts the highest accuracy (84.07%), wekaDeeplearning4j excels in identifying high-risk individuals via better AUC and PRC area, valuable for prioritizing early intervention despite slightly lower overall accuracy (81.85%). Conversely, MLP's low mean absolute error indicates high precision in individual case prediction, ideal for personalized treatments. However, tradeoffs exist: wekaDeeplearning4j requires longer training times, and MLP's precision may sacrifice sensitivity. Choosing the optimal algorithm depends on context and priorities. High accuracy and speed favor RBFN, while superior high-risk identification or precise individual predictions favor wekaDeeplearning4j or MLP, respectively. Understanding these trade-offs is crucial for maximizing deep learning's effectiveness in cardiovascular disease risk prediction.