The repo contains a novel Hybrid Feature Selection method that reduce redundant features to improve the prediction accuracy of Machine Learning Classifiers. The method combines some traditional Feature Selection techniques to build a new one that can overcome the existing drawback of traditional. We named the method WFE. There WFE represents Wrapper, Filter, and Ensemble respectively.
We have implemented the method on a healthcare informatics system and found better outcomes than traditional. Implementing the method our team achieved 100% Accuracy, F1-score and AUC score during the evaluation phase, which made the system top accurate disease detector than previous systems. The method excluded almost 67% less important features from a medical dataset that may help researchers to reduce lab testing costs and time during the early diagnosis of a disease. Currently, the method is under the development stage of our research team.
