Research On Behavior Recognition Of Smart Phone Users Based On Machine Learning

With the continuous development of information technology in recent years,the research of artificial intelligence technology in the direction of smart devices has become a new trend.The research content of this paper is mainly carried out from two aspects: the first aspect is to build a machine learning model training platform at the front end,and the second aspect uses a variety of machine learning models at the bottom to study the behavior recognition of smartphone users,and compare behavior the accuracy rate of recognition under different models.The machine learning model training platform mainly implements functions such as data template download,data file upload,and result file management.The data is put into the underlying algorithm through the front-end platform to realize the training of the model.In the experimental part of the underlying algorithm of user behavior recognition research,a total of five schemes are designed.Scheme 1 uses the LightGBM algorithm to conduct behavior recognition research,and determines the implementation parameters of the entire algorithm by drawing a learning curve.Option two uses Convolutional Neural Networks(CNN)algorithm for behavior recognition research.In order to ensure that the dimensions of the input data are the same each time,an improved Markov Chain Monte Carlo(MCMC)algorithm is used to conduct the entire experiment.The data is resampled,and the convolution structure is improved.The BN(Batch Normalization)layer,residual neurons,and new activation functions are added to the convolution layer.The accuracy of scheme two is lower than scheme one.The experimental data studied in this paper comes from the time series data collected by the smartphone acceleration sensor.Considering that the experimental data is collected in a certain time series,the Long Short Term Memory(LSTM)is compared with the commonly used cyclic neural network.The(Recurrent Neural Network,RNN)network has long-term memory,so in the third scheme,the parallel LSTM network is used for behavior recognition research.The accuracy of scheme three is higher than scheme two and scheme one.Scheme 4 uses the stacking method to combine the two models of LSTM and LightGBM for research,and the accuracy of scheme 4 is higher than that of the first three schemes.Scheme 5 is improved on the basis of the long-term recurrent convolutional neural network(LRCN)model,and proposes the I-LRCN(improve-long-term recurrent convolutional neural network)model,which is compared with the first four The traditional models in the scheme were compared,and the results showed that the recognition accuracy of the I-LRCN model was the highest,and the convergence speed was higher than the scheme 1,2,and 3 models,and lower than the scheme 4model,and the recognition time was the longest among the five schemes.