Published Paper Details:

Preventing privacy and personal information leaks has become a difficult job as smartphone usage has grown quickly. Impersonation is one of the main effects of such leaks. Due to the inability of current preventative measures (such as passcodes and fingerprints) to continually monitor usage and determine if a user is permitted, this kind of unlawful usage is practically hard to stop. Unauthorized users would have complete access to the devices after they were able to circumvent the first security measures, including utilising passwords they had previously stored to get into valuable websites. We introduce KOLLECTOR, a brand-new framework for detecting impersonation that uses a multi-view bagging deep learning method to gather data on successive tapping on a smartphone's keyboard. For continuous user authentication while typing, we build a sequential-tapping biometrics model. We assessed statistically. The results of our comparison between our model and widely established shallow machine techniques show that our system outperforms other methods and can reach an 8.42% equal error rate, a 94.24% accuracy, and a 94.41% H-mean using only the accelerometer and five keyboard taps. We also test the method with only three keyboard taps and discover that it still produces excellent accuracy while providing more possibilities to make more decisions that might lead to more accurate final judgments.