ElGamal Homomorphic Encryption with SMOTE for PET in Occupancy Monitoring by XGBoost

Room occupancy monitoring is an integral part of smart buildings because it can control devices such as air conditioners (AC) and lighting, which increase efficiency and user comfort. However, room occupancy data collected by in-door sensors is personally identifiable information (PII) related to an individual that can reveal the individual's identity. Our research aim is to apply ElGamal homomorphic encryption (HE) as a privacy-enhancing technology (PET) in room occupancy monitoring and evaluate its performance in prediction and privacy preservation. First, we get a dataset from Kaggle for room occupancy monitoring. Then, we assess the accuracy loss of ElGamal HE by comparing four machine learning methods: extreme gradient boosting (XGBoost), random forest, decision tree, and logistic regression. We also address the imbalanced dataset problem of room occupancy monitoring by comparing three oversampling methods: random over-sampling (ROS), synthetic minority oversampling technique (SMOTE), and SMOTE with edited nearest neighbor (SMOTE-EEN). We benchmark the privacy-preserving performance of ElGamal HE with Naccache-Stern HE using two metrics: Kolmogorov-Smirnov (K-S) test and Kullback-Leibler (KL) divergence. Our test results show that XGBoost has the best accuracy on ElGamal HE ciphertext, compared to the random forest, decision tree, and logistic regression, with a value of 0.95. Then, XGBoost also has the smallest accuracy loss, which is 5.12%. Then, SMOTE has the best average recall compared to ROS and SMOTE-ENN, which is 85.29%. Finally, the ElGamal HE ciphertext has a higher KL divergence than Naccache-Stern HE, with a value of 25,309.39 and a K-S test showing a significant difference with the plaintext. The conclusion is that ElGamal HE is the optimum PET for privacy preservation in room occupancy monitoring.