Predicting gender from footfalls using a seismic sensor

This paper deals with gender classification by analyzing footfalls captured by seismic sensor. Gender classification plays an important role in developing different applications like guided navigation system that helps the customer to find different shops in a big shopping mall or to find different departments in a hospital on the basis of their gender. We have tested different classifiers (SVM-Linear, SVM-RBF, Artificial Neural Network, Linear Discriminate Analysis, Logistic Regression and Naive Bayes) on our dataset, consisting of a total 8 hours of seismic data of male and female subjects. We achieved an accuracy and F1 score of 94.56% and 94.56% respectively for SVM Linear classifier. Our event detection technique locates a seismic event and extracts only the corresponding footsteps from the signal. It reduces the feature extraction time by 50%. We have also reported an experimental realization of a real time gender classification system. Experiments were carried out to find the minimum number of samples required to train the machine learning algorithm by maintaining a respectable prediction accuracy.