Feature Selection by LFPSOLF for Enhanced Disease Detection

The accurate diagnosis of diseases is crucial for the treatment and life-saving of patients. Often, the curse of dimensionality in medical data causes incorrect detection of fatal diseases and thereby denies timely treatments leading to higher death tolls. Processing the relevant features alone can achieve improved accuracy at reduced computational cost. Therefore feature selection is highly significant in diagnostic systems and, in line with that, we propose a swarm intelligence-based feature selection method, Leader-follower PSO with Levy Flight (LFPSOLF), adopting the leader-follower concept. The leader performs Levy flight on a random probability basis and the followers do so when their distance from the leader exceeds a threshold, set by the user. The algorithm optimizes the feature set by minimizing the fitness function, the support vector machine (SVM) prediction weighted error. Levy flight of the leader enables the particles to explore the search space and the normal updating of particle position for exploitation. The proposed method was tested on three representative medical datasets and achieved up to 8% improvement in accuracy, while at 40-82% reduction of features. Compared to a few relevant feature selection techniques, the LFPSOLF demonstrated superior accuracy.