Heart Disease Prediction Using Optimized Feature Selection and Classification Techniques

Abstract

Machine learning (ML) in healthcare is gaining popularity, particularly for enhancing the accuracy and timeliness of diagnoses. Machine learning can forecast diseases by analysing enormous volumes of medical data, providing patients and medical professionals with the knowledge they need to make informed decisions regarding disease prevention. Heart disease prediction is crucial as early detection of the condition may save lives and expenses. Therefore, this study aims to extract the most significant variables from a high-dimensional dataset that aid in precisely and accurately classifying heart related disease. The authors of this work propose a hybrid heart disease prediction model by integrating AdaSyn, Particle Swarm Optimization, and Machine Learning techniques. This work uses the Cleveland dataset to assess and validate the system's performance. Firstly, preprocessing of the dataset is accomplish by min-max normalization to standardize the dataset. Thereafter, an Adaptive Synthetic Sampling Approach (ADASYN) was used to balance the dataset.  Furthermore, to select optimal features, Particle Swarm Optimization (PSO) was used.  Lastly, seven different ML models are trained on both full and optimized feature subsets as experimental analysis inputs. The proposed model outperformed conventional models with 91.8% accuracy, 90% precision, 92.8% sensitivity, 91.16% F1 measure, 90.9% Specificity, and 92% Area under the Receiver Operating Characteristic Curve. The classification results demonstrated the strong influence of pertinent characteristics on classification accuracy. When comparing models trained on the whole feature set to those trained on a smaller number of features, the classification models' performance increased noticeably with less training time.