Activity recognition for security personel using machine learning models with smartphone data

In our country, security personnel work in difficult conditions such as terrain, open or closed environments. Ensuring the safety of personnel while on duty is one of the most important principles. In this context, it is possible to monitor the personnel with different technologies while on duty. In this study, assistive technologies that will enable the monitoring of the activities of security personnel during the task are examined. For this, the possibilities of activity recognition were investigated by using sensor data on smart phones and machine learning methods. In the research, sensor data obtained from 30 participants aged between 19-48 and carrying a smart phone were used. 6 different activities were obtained from each participant: walking, climbing, descending, sitting, waiting, reaching. The data were used in models created with Support Vector Machine (SVM), Random Forest (RF), Nearest Neighborhood (KNN), Multilayer Perceptron (MLP) algorithms used in machine learning. Obtained classification results are given in tables together with metrics such as sensitivity, specificity and accuracy. The Receiver Operating Characteristic Curve (ROC Curve) was plotted for each model to monitor data consistency. In order to examine the differences in the classification results, the data set was divided into 90%, 80%, 70% training and 10%, 20%, 30% test data, respectively, and evaluated in the models. The results for the training and test data are shown with confusion matrices and in tabular form.