Improving PIR Sensor Network-Based Activity Recognition with PCA and KNN

With the use of low-cost passive infrared (PIR) sensors in detecting movement, forming a wireless sensor network (WSN) combined with activity recognition (AR), activities or movements that exist in each room can be detected and can be used for health, home automation, and security purposes. Other studies have proven that the hierarchical hidden Markov model (HHMM) method, an a posteriori method is more accurate than unsupervised classification methods such as Naïve Bayes but in another study, unsupervised methods such as k-nearest neighbors (KNN) can show high performance because previously, the datasets go through pre-processing steps. The purpose of this study is to improve the performance of PIR sensor network-based AR using PCA as a pre-processing method and compare the performance with AR in previous studies. In addition, KNN is used as the classification method for AR. To do that, a PIR sensor network needs to be built. 4 PIR sensor nodes are used throughout a test environment house. There are 37150 data that has been collected from all PIR sensors stored in a span of 21 days to build the KNN model. The accuracy results obtained from the KNN model for AR classification is 0.94. The PCA-KNN proposed in this research proves to have higher performance than other studies that also implement AR with PIR sensor network. The proposed method is also a low-cost solution compared to other studies that also implement AR but with more complex sensor combinations.