Research And Implementation Of Identity Recognition Technology Based On Keystroke Dynamics

With the development of Internet technology,people have higher requirements for information security.Existing identification methods,including passwords,passwords,etc.,have not met people's needs,and password theft is not uncommon.So there are some new biophysical-based identification methods,such as fingerprints,voiceprints,irises,etc.Although these biometric technologies have been applied in real life,these technologies are more or less flawed.For example,the finger film that appears today can crack this fingerprint recognition method.Moreover,these biometric technologies require a large amount of expensive equipment to assist,which is not conducive to large-scale popularization.And the identification technology based on keystroke dynamics makes up for this defect.The purpose of this paper is to find an algorithm with higher precision and shorter training time based on previous research.The design experiments in this paper prove that the isolated forest algorithm has higher accuracy(FAR,FRR,AUC)than other anomaly detection algorithms.Among them,the AUC reaches a maximum of 0.98,and the training takes about 150 ms.At the same time,this paper improves the isolated forest algorithm.In the prediction stage of the isolated forest algorithm,the penalty term is added to the normalized anomaly score.Experiments show that the improved algorithm is higher in accuracy than the original algorithm.This paper also analyzes three important parameters of the isolated forest algorithm,namely the number of subsamples,the number of subtrees,and the threshold of abnormal scores.The design experiment analyzed that the abnormal score threshold has the greatest influence on the accuracy of the algorithm when the number of samples is relatively small.In terms of feature engineering,the paper also generates average time feature and standard deviation time feature based on the original features.Experiments show that the accuracy is further improved after increasing the average time feature and standard deviation time feature.This paper also builds a keystroke-based identity recognition system based on improved algorithms including data acquisition,feature processing,model training and prediction,and self-learning.The system designed in this paper is based on the B/S architecture,that is,the front-end browser collects the user's key time series samples,the back-end model training and sample prediction.At the same time,according to the phenomenon that the user's key rhythm changes with time,this paper adds post-processing steps of selflearning,which can capture the changes of user's key habits in time.At the same time,the system has been tested.The results show that the system can meet the requirements in terms of accuracy and timeconsuming,and the user experience is good and useful.