Gait-based Gender Classification Using Smartphone Accelerometer Sensor

People gender and activities recognition are becoming a hot topic in our daily applications through gait information. The very well-known applications are safety-health, security, entertainment and billing. Numerous data mining and machine learning algorithms have been proposed for such issue. Equally, many technologies could be used to observe the people activities to identify their gender and activities. However, the existing solutions and applications suffer from privacy and cost to deploy the solution and their obtained accuracy. For example, when the CCTV camera or Kinect sensors technology are used to identify people, such technologies will violate the privacy since most of the people do not want to take their pictures or videos during their daily activities. Therefore, to tackle such issue, this paper presents a new scheme to identify the gender of the people via onboard Smartphone sensors including accelerometer sensor. Such a scheme requires little interaction with the people; individuals would simply have to hold his/her smartphone and walk normally. Four different data mining techniques and machine learning algorithms are used to identify people gender including: Decision Tree (DT), Support Vector Machine (SVM), k-Nearest Neighbor (k-NN) and Deep learning algorithm (recurrent-neural-network long-short-term-memory ‘RNN-LSTM’). Further, a set of experiments are conducted via Android-based smartphones (to read smartphone accelerometer sensor) and MATALB-2018a packages used to perform the validity of the scheme. The obtained results from the experiments show that the accuracy of the gender identification is about 94.11% via deep learning algorithm (RNN-LSTM) and is around 83.75% by using DT algorithm.