Algorithm Based on KNN and Multiple Regression for the Missing-Value Estimation of Sensors

Abstract

Missing sensor data are unavoidable when sensors are used to monitor a system. These missing data largely affect the sensor applications. When missing data exist, the best method is estimation. Herein, we introduce the k-nearest neighbor on multiple-regression algorithm (KMRA), which builds on the K Nearest Neighbor (KNN) and multiple regression. In the process of estimation, KMRA considers both spatial correlations from its neighbor sensor and time correlations from its own time serials. After computing these two correlations, the algorithm combines them into a unified result of estimation. As KMRA involves spatial and time correlations, it has the efficiency and practicability of an algorithm. Examination results show that KMRA can precisely estimate the missing data.