Pattern Recognition of Human Activity Based on Smartphone Data Sensors Using SVM Multiclass

Abstract

Mobile devices are increasingly sophisticated while smartphones continue to make the latest generation that immerses the supporting tools needed in everyday life such as cameras, GPS, Microphones, and various sensors such as light sensors, a direction sensor, acceleration sensor (i.e., accelerometer) and the gyroscope sensor. This study aims to classify human activities from the accelerometer and gyroscope sensors on a Sony z3+ smartphone. To implement our system, we collect labeled accelerometer and gyroscope data from eight users when they carry out daily activity. Every activity was recorded for 22 seconds, total data that we use every activity is 2000 data with the total amount of data is 16000 data. This data we classify using the Multiclass Support Vector Machine (SVM) method reaches 97.40% accuracy using a 70% ratio as training data and 30% as test data, the classification process takes 5 seconds to classify the data.