使用機器學習預測 3 個電信參數: 地鐵乘客行程

Abstract

none

This study focuses on the use of machine learning techniques to predict three important telecommunication parameters in the context of metro passengers itineraries. These parameters are the base station changes, the latency of the signal and the number of packet loss. They represent abnormal phenomenon, or events that can alter the phone’s performance. Being able to successfully predict it can lead to a better anticipation of these issues and enhance the user’s experience.

The primary objective of this research is to compare different machine learning models for real-time predictions of the studied parameters. To do so, different algorithms (Neural Networks, Recurrent Neural Networks, LSTM and ARIMA), as well as several sets of features will be used. A comparison on the error metrics will also be conducted. The study took novel approaches in the nature of the studied parameters and the prediction delay, as it aims to forecast the value a few seconds into the future. Additionally, it proposes new solutions to make predictions with a dataset mainly composed of zeros.

Overall, this study contributes to the understanding of machine learning applications in predicting telecommunication parameters in the case of metro passengers itineraries. The findings suggest that the selected machine learning algorithms, combined with appropriate error metrics and innovative approaches, offer reliable solutions for real-time predictions for the three studied parameters. Besides, the proposed solutions to avoid models that always predict zero with datasets mainly composed of null values proved to be successful. The predictions from the chosen models will help in the decision-making of the settings of the phone, to avoid abnormal phenomenon and maintain a good performance throughout the user’s route.