A Balanced Expert System to Manage Parkinson's Disease by Identifying Major Risk Factors: B-TESM-PD

Abstract

With changing lifestyles and emergence of modern ways of living, people fail to take enough care of their physical health and increase their chance of falling prey to fatal and non-fatal diseases. With tremendous advancement in science and technology, it is expected to have a system that can help to manage this problem. One of such prevailing diseases is Parkinson’s disease (PD). The part of the brain that controls movement, posture, and also emotional imbalance is affected in PD. The symptoms of PD may differ from one individual to another. PD is detected basically in elderly people but now it has turned out to be a chronic disease affecting all age groups. As of its severity it is better to have an expert system which can identify the major risk factors of PD in the early stage and accordingly some measures may be taken to control those risk factors. It would, therefore, be really a great help to the world to have one such expert system for managing a PD. Therefore, this paper proposes an expert system named balanced-transparent expert system for managing-PD (B-TESM-PD) that can identify the major risk factors by generating rules from decision tree (DT). B-TESM-PD encompasses of five phases: preprocessing, rule generation, rule selection, rule pruning and merging, and risk factor identification. The imbalance nature of the PD dataset is treated by the preprocessing stage using a dominant undersampling technique, Tomek Link. Rule generation phase generates rules using decision tree. Rule selection stage selects the transparent rules, rule pruning and merging removes the redundant and inefficient rules from the rule set and then merges the pruned rule set to a single rule, and lastly the major risk factor of PD is identified using the risk factor identification phase. The model is validated with the PD dataset collected from UCI repository. Standard decision tree performance is compared with the proposed model.