Mobile communication has become a vital part of modern communication. The cost of network infrastructure has become a deciding factor with rise in mobile phone usage. Subscriber mobility patterns have major effect on load of radio cell in the network. The need for data analysis of subscriber mobility data is of utmost priority. The paper aims at classifying the entire dataset provided by Telenor, into two main groups i.e. Infrastructure stressing and Infrastructure friendly with respect to their impact on the mobile network. The research aims to predict the behavior of new subscriber based on his MOSAIC group. A heuristic method is formulated to characterize the subscribers into three different segments based on their mobility. Tetris Optimization is used to reveal the “Infrastructure Stressing” subscribers in the mobile network. All the experiments have been conducted on the subscriber trajectory data provided by the telecom operator. The results from the experimentation reveal that 5 percent of subscribers from entire data set are “Infrastructure Stressing”. A classification model is developed and evaluated to label the new subscriber as friendly or stressing using WEKA machine learning tool. Naïve Bayes, k-nearest neighbor and J48 Decision tree are classification algorithms used to train the model and to find the relation between features in the labeled subscriber dataset

Context With the rapidly growing demand for cellular networks’ capacityand coverage, effective planning of Network Infrastructure (NI) has been amajor challenge for the telecom operators. The mobility patterns of different subscriber groups in the networks have been found to be a crucialaspect in the planning of NI. For a telecom operator, it is important to havean estimate of the efficiency (in terms of the Network Capacity - numberof subscribers that the network can handle) of the existing NI. For thispurpose, Lundberg et. al., have developed an optimization based strategycalled as Tetris Strategy (TS), based on the standard subscriber groupingapproach called MOSAIC. The objective of TS is to calculate the upperbound estimate of the efficiency of the NI. Objectives The major objective of this thesis is to compare the efficiencyvalue of the NI when the subscribers are grouped (clustered) based on theirmobility patterns (characterized by a mobile trajectory) with the efficiencyvalue obtained when the subscribers are grouped based on the standardsubscriber grouping approach - MOSAIC. Methods Literature Review (LR) has been conducted to identify the stateof the art similarity/distance measures and algorithms to cluster trajectory data. Among the identified ones, for conducting experiments, LongestCommon Subsequences has been chosen as a similarity/distance measure,and Spectral and Agglomerative clustering algorithms have been chosen.All the experiments have been conducted on the subscriber trajectory dataprovided by the telecom operator, Telenor. The clusters obtained from theexperiments have been plugged into TS, to calculate the upper bound estimate of the efficiency of the NI. Results For the highest radio cell capacity, the network capacity valuesfor Spectral clustering, Agglomerative clustering and MOSAIC groupingsystem are 207234, 148056 and 87584 respectively. For every radio cellcapacity value, the mobility based clusters resulted in a higher network efficiency values than the MOSAIC. However, both spectral and agglomerativealgorithms have generated a very low quality clusters with the silhouettescores of 0.0717 and 0.0543 respectively. Conclusions Based on the analysis of the results, it can be concluded that,mobility based grouping of subscribers in the cellular network provide highernetwork efficiency values compared to the standard subscriber grouping systems such as MOSAIC.