Early Identification Of Heart Disease Using An Enhanced Data-Driven Predictive Approach

Heart disease is among the major killers of people around the world and therefore there is the motivation of early and accurate diagnosis. This paper proposes a DL method of predicting HD using the UCI Cleveland HD Dataset. The proposed approach is based on a three-stage pipeline, that is, data preprocessing, FS, and classification. Data preprocessing is performed by the IQR method to eliminate outliers; the FS step is performed using the CST to retrieve the most relevant categorical attributes; and three different DL models, the RNN, ANN, and CNN will perform the classification stage. These are the main metrics determining the performance of each model, accuracy, precision, recall, and AUC-ROC. Tested results demonstrate that the model of RNN along with IQR-based preprocessing and CST-based feature extraction yield better results than the ANN and CNN models. In particular, the RNN provides better results with higher accuracy and better values of recall and AUC-ROC, thus being more successful in minimizing the false negatives which is important in medical diagnosis. This technique shows the capabilities of DL methods, especially RNNs, in creating stable predictive models of CVD, thus helping medical workers to intervene and prepare a treatment plan at an early stage.