Analysis of Mobile Network Data Using Unsupervised Learning

Unsupervised Learning techniques in machine learning (ML) have attracted recently the interest of mobile network industry. They show a great potential in helping overcoming the tremendous challenges facing mobile operators in their network management. One of the issues is the management of radio resource. In LTE technology, which utilizes OFDMA technique, power resource block (PRB) is of prime importance. For the future mobile generation, the ever-increasing traffic and the diversity of services will put enormous strain on radio resources. In order to tackle such problems some unsupervised learning methods such as clustering may provide some help. This paper analyses a real mobile dataset provided by a national mobile operator. It tries to gain an insight into the mobile data and perform adequate exploration for possible patterns, trends and similarities among a set of LTE cells. The dataset collected consists of 82 LTE cells corresponding to ten sites distributed among six cities in Algeria. The results of the exploratory data analysis (EDA) provide useful insights into the mobile dataset. Furthermore, unsupervised learning clearly identifies five distinct clusters with acceptable clustering metrics. This paper employed a methodology that integrated LTE domain expertise and machine learning. The efficacy of the feature engineering process could be enhanced in future work by leveraging additional dataset features provided by the mobile network operator (MNO). A significant obstacle in the analysis of MNO datasets is the limited availability of data, primarily due to privacy regulations and commercial interests.