Healthcare is the domain which is indispensable for leveraging human health and also minimize mortality rate caused by different diseases. One such disease is brain stroke. Medically brain stroke is the condition that occurs due to poor blood flow to brain causing cell death and it is causing millions of deaths all over the world. According to WHO in 2019, stroke caused more than 6 million deaths across the globe. There are many machine learning methods used for stroke detection using data driven approach. However, their performance is deteriorated when the training data quality is mediocre. To overcome this problem, some feature selection methods came into existence. Those methods could improve performance of prediction models. Nevertheless, there is still need for leveraging prediction performance. In this paper, we proposed a hybrid feature selection method to enhance accuracy of prediction of stroke from the given benchmark dataset. Different prediction models are used along with the proposed hybrid feature selection method along with other existing feature selection methods. The experimental results revealed many useful insights. First, all brain stroke prediction models could perform well with feature selection methods when compared with the models without feature selection. Second, the stroke prediction models exhibited highest performance with the proposed hybrid feature selection model. Thus the proposed stroke detection framework can be used in Clinical Decision Support Systems (CDSSs).