INDOOR OCCUPANCY EVALUATION WITH ADVANCED OPTIMIZED SUPPORT VECTOR MACHINE (SVM)

Abstract

Indoor occupancy estimation is crucial in energy management, security and building automation systems. The Machine learning approaches that are normally used to improve the reliability and accuracy of this estimate face challenges in capturing dynamic occupant behavior in diverse environments. This research develops an optimized Support Vector Machine (SVM) learning model using Pelican Optimization Algorithms (POA) to enhance accuracy and efficiency in estimating indoor occupancy. The model was developed using an occupancy dataset, organized, cleaned and pre-processed. Support Vector Machine with a Pelican Optimization Algorithm (POA-SVM) and Support Vector Machine (SVM) were used as classifiers for feature extraction and classification, learning and classifying unsupervised or supervised and Hyperparameter tuning was used to optimize parameters, enhancing generalization and capturing complex relationships in data. The result shows that the optimized SVM model performs better than the earlier employed machine learning approaches and the ordinary support vector machine in accuracy and precision by 99%. It also performs better in terms of False Positive Rate (FPR), recall and specificity. The algorithm has been used as a low-cost people-counting strategy for real-time applications, making it a valuable tool for estimating indoor occupancy in diverse environments.