@article { author = {Gandomkar, Muhammad and Sarang, Reza and Gandomkar, Ziba}, title = {A Framework to Identify and Count Popular Exercises Using Smartphone Sensors Based on Machine learning}, journal = {Journal of Advanced Sport Technology}, volume = {4}, number = {2}, pages = {29-37}, year = {2020}, publisher = {University of Mohaghegh Ardabili}, issn = {2538-5259}, eissn = {2783-3399}, doi = {}, abstract = {Smartphones have wide range of sensors such as gyroscopes or inertial sensors, which can be used for recognizing and tracking exercises. A framework, called TrainingPal, was proposed to automatically identify five types of cardio exercises and five types of resistance exercises. Included exercises were running, walking, rowing, using elliptical machine, and jumping jack. Sit-up, bench dip, push-up, squat, and lunge were included as popular resistance exercises. In addition to recognition of each exercises, the proposed framework was able to count number of repetitions of each exercise. To train and test the proposed framework, data was collected from Samsung Galaxy S7 edge, which was attached to the outer side of arm approximately 10 to 12 cm below the shoulder. To avoid overfitting, we used leave-one-subject-out cross validation. An overall accuracy of 91.71% was achieved in identifying different types of exercises. The accuracy ranged from 100% for push-ups to 60.33% for bench dips. The accuracy of the proposed framework in counting the exercises was 90%. The results suggested that the proposed framework can be used for identifying and tracking of the included exercises. The framework can be extended to other wearable devices. }, keywords = {Exercise recognition,Exercise tracking,Inertial sensors,Smartphone}, url = {https://jast.uma.ac.ir/article_950.html}, eprint = {https://jast.uma.ac.ir/article_950_b777e6bed710f42cf22909958b80d777.pdf} }