Unintended Notification Swipe Detection System

Users often make errors in touch or swipe while interacting with mobile phones. One of the common areas of concern is accidental swiping off important notifications. We show that these unintentional notification swipes can be accurately detected by using simple touch and swipe features recorded while performing the gesture. The pre-installed touch and grip sensors were used to record data of 20 different participants asked to perform intentional and unintentional touch gestures. The various features taken into account are extracted from user's hand movement on the screen and by identifying single-handed or two-handed grip. In addition to three previously published features- Touch Time, Swipe Velocity and Average Touch Size, we introduce three novel features in our system namely Swipe Stretch, Nearest Edge Gap based on grip and Notification Expansion Action. We trained our model using Random Forest (RF) classifier and Neural Networks (NN) and achieved the accuracy of 98.8% and 100% respectively. The results prove that the model can successfully detect unintentional notification swipe and touch gestures in real time. The novelty of our research lies in considerable improvement of accuracy over previous published works attributed to a larger feature set inclusive of proposed features.