Independent Components In Telecommunication Call-detail Records
Peter Laszlo Lakatos, B. Egri, Z. Cseh · WIT transactions on information and communication technologies · 2003
Although Independent Component Analysis (ICA) has gained a recognized place among neural network models [l], data mining applications have been limited, especially for marketing purposes. In our paper we will present a unique ICA application, in which ICA outperforms other data mining and statistical methods. We've applied ICA to the call-detail records of a fixed-line telecommunication company's customers. Telecommunication companies can observe the duration, time and tariff zone of the incoming/outgoing calls, which, in this context, represent the mixed input signals. These records are generated as a result of a (supposedly linear) mixture of independent signals representing the telecommunication needs of a family, like chatting with friends, business calls or Internet usage. The aim was to disclose these underlying independent components. These components can be interpreted according to the subscribers' independent telecommunications needs, potentially the most useful information for decision makers as it accurately describes overall customer behavior. ICA can show which components are most important for each customer. The insights gained through these methods create new opportunities for identifying important market segments and facilitating customer demand-based product development and is an effective way of dealing with churn in the telecommunication realm. In this paper we give a short introduction to the main features of the ICA method, focusing on its application in the telecommunications marketing area and briefly summarizing the results of our work on real data.