Homomorphic Encryption for Privacy Preservation in Occupancy Sensor-Based Smart Lighting

Occupancy sensor-based smart lighting is very practical because the light status can change automatically according to human activity under the lighting area. However, on the other hand, the user occupancy displayed by the occupancy sensor is sensitive personal information. Our research aim is to apply homomorphic encryption to occupancy sensor data in smart lighting and evaluate its performance. In the first step, we obtain occupancy data for smart lighting using classification integrated moving averages (CIMA). We then design and apply homomorphic encryption to the data. We use three machine learning models to test the data utility of this method: k-nearest neighbors (KNN), adaptive boosting (AdaBoost), and decision trees. Meanwhile, for the privacy preservation benchmark, we use differential privacy. Several metrics to measure the performance of this privacy-preservation mechanism are information loss, entropy, kernel density estimation (KDE), Kullback-Leibler (KL) divergence, and Earth mover's distance (EMD). The test results show that AdaBoost provides the lowest information loss compared to the other two methods, namely 6%. This shows that homomorphic encryption can maintain data utility. Compared with differential privacy, homomorphic encryption has a KDE with low kurtosis. Then, the KL divergence, EMD, and average entropy values of homomorphic encryption are also lower than those of differential privacy. Homomorphic encryption provides optimum privacy preservation in occupancy sensor-based smart lighting.