Call Frequency Time Series Forecasting Based On LSTM

Call frequency time series are extracted from Call Data Records and can be used to explore human activity patterns and predict call activity. In practical applications, accurate prediction of call frequency is very important, and can be used for optimizing communication network resource allocation, improving business processes and other aspects. In this paper, we design the network structure of LSTM (Long and Short Term Memory Network) and optimize the key parameters of the model to determine the optimal LSTM model to predict short term call frequency. In the experiment, we extracted 10 real call records for verification, and use these time series to test the model we designed. The average MAPE (Mean Absolute Percentage Error) is 5.721%. The results indicate that the model we used is a fast and effective one, which is advantageous for predicting call activity in future practical work.