MFES Framework for Efficient Feature Selection Among Subsystems in Intelligent Building

The increasing trend of problem representation and high-dimensional data collection calls for the utilization of feature selection in many machines learning tasks and big data representations. However, identifying meaningful features from thousands of related features in the smart home data which are dissimilar in nature remains a nontrivial task. This has prompted for the deployment of a feature selection algorithm (FSA) that provides two possible solutions. First, to provide an efficient scheme that best optimizes the features for subsystem decisions and second, tackles feature subset selection bias problem. In this paper, a MFES framework for feature selection is proposed that uses a hybrid mechanism to tackle the problem of feature subset selection bias in intelligent building data. The mechanism uses the effectiveness of filters and accuracy of wrappers to obtain significant features for prediction. The proposed MFES framework resulted in 92.17% of accuracy as compared to the baseline approach resulting in 87.21% of accuracy. The experimental results show that efficient and better prediction accuracy can be achieved with a smaller feature set.