An intelligent recommender system using machine learning association rules and rough set for disease prediction from incomplete symptom set

Digital devices are an integral component of the healthcare sector. With the advancement of modern technology with Artificial Intelligence (AI) and Machine Learning (ML), an automated diagnosis system with promising results is not a difficult task. This study aims to develop a recommender system (RS) for better diagnosis and improvement of patient care by hybridizing machine learning association rules (AR) and rough set theory (RST) to classify acute and life-threatening diseases. Initially data is preprocessed using binary, on-hot vector, and min-max scale to remove the noise. RST is used for feature selection to deal with incompleteness, inconsistency, and vagueness. We have designed an Associated Symptom Selection (ASS) algorithm to extract the mutually associated symptoms which need to be further matched in the existing database for prediction. ASS is especially helpful in detecting neurodevelopmental type diseases because the symptoms are usually not detectable by standard tests, and observations of behavioral expressions do general testing. The experiment is carried out using six popular ML classifiers such as AR, Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN), Linear Support Vector Machine (LSVM), and Naive Bayes (NB) on a publicly available datasets. Performance was compared among different classifiers regarding the accuracy, precision, recall, F1-score, and J-Score value. The experimental result shows that AR performs better on clinical data with an accuracy of 94.40%, precision of 90.73%, recall of 94.45%, F1-score of 92.55%, and J-score of 95.14% and on autism with 98.7% accuracy, 98% precision, 97.8% recall, 97.9% F1-score, and 97.12% J-score respectively.