Gender and Intent Classification From Finger Swiping Behaviours on Gesture Keyboards Using LSTM

Abstract

With the increasing number of smartphone users worldwide, gender prediction research has shifted from using keyboard strokes to touch behaviour from smartphone. One of the touching behaviours is swiping behaviour from gesture keyboard. The goal of this research was to produce a classification result for gender and intent (Quickly, Accurately, Creatively) by utilizing finger swiping behaviors on gesture keyboards. Research methods were conducted by data development, data training, data testing and performance evaluation. Long Short-Term Memory (LSTM) recurrent neural network architecture served as the main structure for data training. Evaluation used cross validation method with performance evaluation metrics (F1 score and Area Under Curve (AUC)). The highest result of gender classification is 0.88 F1 score and 0.84 AUC for male and 0.83 F1 score and 0.85 AUC for female. As for intent classification it is 0.76 F1 Score and 0.91 AUC. It can be concluded that gender and intent can be classified using LSTM architecture from finger swiping behaviors.