PIF dataset: a comprehensive dataset of physiological and inertial features for recognition of human activities

Activities and falls monitoring systems using wearable technology have a promising future. The publicly available datasets are based on a few inertial features only acquired with an accelerometer, gyroscope, smartphone or smart Watches. The activities and falls performed are also less. In this study, a dataset is created by collecting physiological features along with inertial features which will help in developing and validating systems studying the effect of physiological features on the detection and prediction of falls and activities. The dataset consists of 7 activities and 8 falls for inertial data; 2 activities for ECG data; 6 activities for EMG data and 6 activities for GSR data. Basic body parameters like height, weight, etc. along with beats per minute, SpO2 and blood pressure are also recorded for 12 subjects. The collected data is analyzed statistically using a boxplot, pair plot, correlation heatmap and p value. The activities are classified using SVM, KNN, RF and DT. For GSR, more than 90% accuracy is achieved and for EMG, the accuracy is less than 80%. For IMU data, more than 95% accuracy is achieved. The results encourage combining inertial, physiological and basic body parameters to detect and predict falls and activities.