Optimization-enabled deep stacked autoencoder for occupancy detection

Occupancy detection is the main building block of residential and commercial building automation systems. Currently, it is difficult to determine when and where people occupy a commercial building. The accurate occupancy detection in the buildings is used to save energy. Hence, this paper presents an occupancy detection approach for detecting the person’s count in the room or building using the proposed Chaotic Whale Spider Monkey (ChaoWSM) + Deep stacked autoencoder. The input data are initially fed to the pre-processing step. The pre-processing is done using missing value imputation and log transformation. Then, the feature reduction is performed from the pre-processed data. Here, the features are reduced based on probabilistic principal component analysis. On the next step, the reduced features are fed to the occupancy detection module where the occupancy is detected on the basis of the support vector neural network classifier. If the occupancy is detected, the features are forwarded to the person’s count identification. The identification of a person’s count is carried out based on deep stacked autoencoder, which is trained by an optimization approach. The optimization is done using the proposed Chaotic Whale Spider Monkey (ChaoWSM) optimization, which is the integration of Chaotic Whale optimization and the spider monkey optimization. The performance of occupancy detection based on ChaoWSM + Deep stacked autoencoder is evaluated based on sensitivity, specificity, accuracy and MSE. The proposed ChaoWSM + Deep stacked autoencoder method achieves the maximal accuracy of 0.945, maximal sensitivity of 0.946, the maximal specificity of 0.949 and the minimal MSE of 0.233 by varying the training data percentage for deep stacked autoencoder.