Feature selection for risk detection of strokes: A 5-year longitudinal study

Stroke disease places a heavy burden on society in the world. Risk detection of strokes is a challenge in medical field. This paper investigates biomedical tests and electronic archives on 792 records (each record includes 28 features), which contain 398 records in the last 5 years before stroke at a community hospital. We proposed a new hyper feature selection model combined support vector machines (SVM) with the glowworm swarm optimization (GSO) algorithm based on STD (Standard Deviation) of features. The results show that the proposed model can achieve 73.23% accuracy by means of the 18 features among the original dataset
¹
.