Design and implementation of efficient data mining models using CDR data for density analysis and prediction

Abstract

A Call Data Record (CDR) is a record that stores information about call details and newlineactivities in a telephonic network especially the mobile network. It contains temporal newlineand spatial data, and can also convey other information that would be helpful to the newlineuser. A large number of vehicles on roads creates substantial traffic, which makes it newlinevery difficult to maintain safety and control traffic especially in the urban areas. newlineSeveral works were carried out in the past to estimate the traffic density. However, newlinethey were inappropriate and quite expensive, owing to the dynamics of the traffic flow. newlineThis thesis proposes the use of CDR data to find the density of a location, and to track newlinethe location of the mobile user, which can be useful to control the traffic at a newlineparticular day and at particular time in applications such as traffic control applications. newlineFrequency Pattern Mining (FPM) and Generalized linear Models (GLM) are used for newlinethe prediction and to find the co-occurrence of the position associated with a mobile newlineuser. Recurrent neural Networks (RNN) using LSTM (long Short-term memory) are newlineused for the time series prediction. A Bacterial Foraging Optimization Algorithm newline(BFO) is also proposed, to tackle the local optimality problem in K-Means clustering newlinetechnique to produce a more cohesive cluster .The algorithm was evaluated over newlinestandard data sets and performance was found to be effective in terms of accuracy. newline