Human Activity Recognition Using K-Nearest Neighbor Machine Learning Algorithm

Smart factory in the era of Industry 4.0 requires humans to have continuous communication capabilities among each other's and with the existing smart assets in order to integrate their activities into a cyber-physical system (CPS) within the smart factory. Machine learning (ML) algorithms can help precisely recognize the human activities, provided that well-designed and trained ML algorithms for high performance recognition are developed. This paper presents a k-nearest neighbor (KNN) algorithm for classification of human activities, namely Laying, Downstairs walking, Sitting, Upstairs walking, Standing, and Walking. This algorithm is trained and the algorithm's parameters are precisely tuned of for high accuracy achievement. Experimentally, a normalized confusion matrix, a classification report of human activities, receiver operating characteristic (ROC) curves, and precision-recall curves are used to analyze the performance of the KNN algorithm. The results show that the KNN algorithm provides a high performance in the classification of human activities. The weighted average precision, recall, F1-score, and the area under the micro-average precision-recall curve for the KNN are 90.96%, 90.46%, 90.37%, and 96.5%, respectively, while the area under the ROC curve is 100%.