Combined Optimization and Modified Performance Metrics for Automated Model and Parameter Selection in Telecom Customer Churn Prediction

DOI

10.2991/itsmssm-17.2017.42How to use a DOI?

Keywords

binary classification, artificial neural network, genetic algorithm, combined optimization, performance metric.

Abstract

- In this paper, we consider the construction of an optimization algorithm designed to identify, in automatic way, the optimal parameters and algorithms for binary classification in the task of customer churn prediction in telecommunication. The currently supported methods for classification include Decision Trees, k-Nearest Neighbors, Support Vector Machines, and Back-Propagation Artificial Neural Networks. It is shown that in certain cases the standard textbook metrics of classification model quality (e.g. accuracy, precision, recall, and AUC) are not descriptive enough. Thus, we evaluate the algorithms using the modified recall metrics: Weighted Recall metric, and Weighted Recall and Run-Time metric. These metrics are appropriate for use in automatic mode without continuous expert supervision. We use genetic algorithms as optimization algorithms. To improve standard implementations, we introduce a combined optimization approach based on D.H. de Vries model of disaster evolution and the island model.

Copyright

© 2017, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Stanislav Alkhasov
AU  - Alexey Tselykh
PY  - 2017/12
DA  - 2017/12
TI  - Combined Optimization and Modified Performance Metrics for Automated Model and Parameter Selection in Telecom Customer Churn Prediction
BT  - Proceedings of the IV International research conference "Information technologies in Science, Management, Social sphere and Medicine" (ITSMSSM 2017)
PB  - Atlantis Press
SP  - 196
EP  - 200
SN  - 2352-538X
UR  - https://doi.org/10.2991/itsmssm-17.2017.42
DO  - 10.2991/itsmssm-17.2017.42
ID  - Alkhasov2017/12
ER  -