Cellular Data Analysis using Machine Learning for Crowd Gathering Prediction

The ability to accurately forecast the volume of cellular traffic is essential to maximise the efficiency with which network resources are allocated, enhancing the user experience and boosting overall network performance. Deep Learning methods like Long Short-Term Memory (LSTM) and Bidirectional LSTM (Bi-LSTM) have garnered much interest recently. They can recognise long-term relationships and temporal patterns in sequential data. This article compares LSTM and Bi-LSTM to predict the cellular traffic load in a cell tower. Metrics such as Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE) are used to compare the two models' respective levels of accuracy. Findings show that Bi-LSTM model beats the LSTM model, particularly when dealing with complicated and non-linear traffic patterns. Further as per forecast value of data traffic in a tower and comparing actual data usage alongwith number of users connected, a smart application is suggested which can predict for outcome as Crowd Gathering or Data Broadcast situation for an hour in a tower coverage area.This can alert local administration to review situation proactively, there can be few exceptions which need to be factored like festival,sports event etc. before reaching a conclusion.