Improved churn prediction in telecommunication industry by analyzing a large network

Customer retention in telecommunication companies is one of the most important issues in customer relationship management, and customer churn prediction is a major instrument in customer retention. Churn prediction aims at identifying potential churning customers. Traditional approaches for determining potential churning customers are based only on customer personal information without considering the relationship among customers. However, the subscribers of telecommunication companies are connected with other customers, and network properties among people may affect the churn. For this reason, we proposed a new procedure of the churn prediction by examining the communication patterns among subscribers and considering a propagation process in a network based on call detail records which transfers churning information from churners to non-churners. A fast and effective propagation process is possible through community detection and through setting the initial energy of churners (the amount of information transferred) differently in churn date or centrality. The proposed procedure was evaluated based on the performance of the prediction model trained with a social network feature and traditional personal features.     

Improving customer attrition prediction by integrating emotions from client/company interaction emails and evaluating multiple classifiers

Predicting customer churn with the purpose of retaining customers is a hot topic in academy as well as in today’s business environment. Targeting the right customers for a specific retention campaign carries a high priority. This study focuses on two aspects in which churn prediction models could be improved by (i) relying on customer information type diversity and (ii) choosing the best performing classification technique. (i) With the upcoming interest in new media (e.g. blogs, emails, ...), client/company interactions are facilitated. Consequently, new types of information are available which generate new opportunities to increase the prediction power of a churn model. This study contributes to the literature by finding evidence that adding emotions expressed in client/company emails increases the predictive performance of an extended RFM churn model. As a substantive contribution, an in-depth study of the impact of the emotionality indicators on churn behavior is done. (ii) This study compares three classification techniques – i.e. Logistic Regression, Support Vector Machines and Random Forests – to distinguish churners from non-churners. This paper shows that Random Forests is a viable opportunity to improve predictive performance compared to Support Vector Machines and Logistic Regression which both exhibit an equal performance.     

Estimating the effect of word of mouth on churn and cross-buying in the mobile phone market with Markov logic networks

Much has been written about word of mouth and customer behavior. Telephone call detail records provide a novel way to understand the strength of the relationship between individuals. In this paper, we predict using call detail records the impact that the behavior of one customer has on another customer's decisions. We study this in the context of churn (a decision to leave a communication service provider) and cross-buying decisions based on an anonymized data set from a telecommunications provider. Call detail records are represented as a weighted graph and a novel statistical learning technique, Markov logic networks, is used in conjunction with logit models based on lagged neighborhood variables to develop the predictive model. In addition, we propose an approach to propositionalization tailored to predictive modeling with social network data. The results show that information on the churn of network neighbors has a significant positive impact on the predictive accuracy and in particular the sensitivity of churn models. The results provide evidence that word of mouth has a considerable impact on customers' churn decisions and also on the purchase decisions, leading to a 19.5% and 8.4% increase in sensitivity of predictive models.     

CRM at a pay-TV company: Using analytical models to reduce customer attrition by targeted marketing for subscription services

The early detection of potential churners enables companies to target these customers using specific retention actions, and subsequently increase profits. This analytical CRM (Customer Relationship Management) approach is illustrated using real-life data of a European pay-TV company. Their very high churn rate has had a devastating effect on their customer base. This paper first develops different churn-prediction models: the introduction of Markov chains in churn prediction, and a random forest model are benchmarked to a basic logistic model.The most appropriate model is subsequently used to target those customers with a high churn probability in a field experiment. Three alternative courses of marketing action are applied: giving free incentives, organizing special customer events, obtaining feedback on customer satisfaction through questionnaires. The results of this field experiment show that profits can be doubled using our churn-prediction model. Moreover, profits vary enormously with respect to the selected retention action, indicating that a customer satisfaction questionnaire yields the best results, a phenomenon known in the psychological literature as the ‘mere-measurement effect’.     

Social network analytics for churn prediction in telco: Model building,evaluation and network architecture

Social network analytics methods are being used in the telecommunication industry to predict customer churn with great success. In particular it has been shown that relational learners adapted to this specific problem enhance the performance of predictive models. In the current study we benchmark different strategies for constructing a relational learner by applying them to a total of eight distinct call-detail record datasets, originating from telecommunication organizations across the world. We statistically evaluate the effect of relational classifiers and collective inference methods on the predictive power of relational learners, as well as the performance of models where relational learners are combined with traditional methods of predicting customer churn in the telecommunication industry. Finally we investigate the effect of network construction on model performance; our findings imply that the definition of edges and weights in the network does have an impact on the results of the predictive models. As a result of the study, the best configuration is a non-relational learner enriched with network variables, without collective inference, using binary weights and undirected networks. In addition, we provide guidelines on how to apply social networks analytics for churn prediction in the telecommunication industry in an optimal way, ranging from network architecture to model building and evaluation.     

SVM在移动通信客户流失预测中的应用研究

使用支持向量机(SVM,Support Vector Machine)数据挖掘方法对移动通信行业客户流失倾向进行预测,对支持向量机同决策树算法预测的结果进行对比,结果表明支持向量机对本文所选取的属性数据具有更强的分类能力,而且在不同训练数据规模情况下预测模型有较好的稳定性。实验证实,运用本研究模型选取全体客户的22.31%,可以预测出50.07%流失的客户,表明本研究中提出的预测模型具有实际应用价值。     

Measuring the success of retention management models built on churn probability,retention probability,and expected yearly revenues

We claim that often marketers have not all the information to develop various marketing campaign models. For example, marketers may have sufficient information to build a model for predicting possible churners, while they may have no clues of which customers are most likely to accept a retention campaign. In this paper, we first show that the information useful for a successful churner prediction model alone is not sufficient to develop a successful retention marketing program. In such a case, we claim that only theory-based simulation approach is feasible. In particular, it is claimed that optimal retention management models should consider not only churn probability but also retention probability and expected revenues from target customers. To validate our claim, we develop and compare five retention management models based on churn probability, retention probability, expected revenues, and combination of these models along with different evaluation metrics. Our experimental results show that the retention management model with the highest accuracy in predicting possible churners is not necessarily optimal because it does not consider the probability of accepting retention promotions. In contrast, the retention management model based on both churn and retention probability is the best in terms of predicting customers who are most likely to positively respond to retention promotions. Ultimately, the model based on expected yearly revenue of customers accrues the highest revenues across most target points, making it the best model out of five churn management models.     

Employee churn prediction

Customer churn is a notorious problem for most industries, as loss of a customer affects revenues and brand image and acquiring new customers is difficult. Reliable predictive models for customer churn could be useful in devising customer retention plans. We survey and compare some major machine learning techniques that have been used to build predictive customer churn models. Employee churn (or attrition) closely related but not identical to customer churn is similarly painful for an organization, leading to disruptions, customer dissatisfaction and time and efforts lost in finding and training replacement. We present a case study that we carried out for building and comparing predictive employee churn models. We also propose a simple value model for employees that can be used to identify how many of the churned employees were “valuable”. This work has the potential for designing better employee retention plans and improving employee satisfaction.     

Including spatial interdependence in customer acquisition models: A cross-category comparison

Within analytical customer relationship management (CRM), customer acquisition models suffer the most from a lack of data quality because the information of potential customers is mostly limited to socio-demographic and lifestyle variables obtained from external data vendors. Particularly in this situation, taking advantage of the spatial correlation between customers can improve the predictive performance of these models. This study compares an autoregressive and hierarchical technique that both are able to incorporate spatial information in a model that can be applied on large datasets, which is typical for CRM. Predictive performances of these models are compared in an application that identifies potential new customers for 25 products and brands. The results show that when a discrete spatial variable is used to group customers into mutually exclusive neighborhoods, a multilevel model performs at least as well as, and for a large number of durable goods even significantly better than a frequently used autologistic model. Further, this application provides interesting insights for marketing decision makers. It indicates that especially for publicly consumed durable goods neighborhood effects can be identified. However, for more exclusive brands, incorporating spatial information will not always result in major predictive improvements. For these luxury products, the high spatial interdependence is mainly caused by homophily in which the spatial variable is a substitute for absent socio-demographic and lifestyle variables. As a result, these neighborhood variables lose a lot of predictive value on top of a traditional acquisition model that typically is based on such non-transactional variables.     

Comparative analysis of artificial neural networks and dynamic models as virtual sensors

This paper presents a comparison of predictive models for the estimation of engine power and tailpipe emissions for a 4 kW gasoline scooter. This study forms a benchmark toward establishing an online emissions control and monitoring system to bring the emissions to within specific limits. Three emissions predictive models were investigated in this study; direct and series artificial neural network (ANN) models and a MATLAB dynamic model. The direct models takes variables lambda, throttle position, engine and vehicle speed to predict the engine power and the emissions CO, CO₂ and HC. The series model first takes the mentioned input to predict the engine power and consequently using the engine power as the fifth input to predict the emissions. For the ANN models, two multilayered networks were compared and analyzed; the backpropagation (BP) and optimization layer-by-layer (OLL) algorithms. The predictive accuracy for each algorithm were compared and it was found that the OLL network is the most accurate with a maximum mean relative error (MRE) of 1.78% and 1.38% for the direct and series predictive model respectively. Comparative results showed that the series neural network model gives the most accurate predictions, with MRE of 0.63% and 0.47% for the engine power and emissions respectively. The series neural network model can be seen as generic virtual power and emissions sensors, substituting costly and cumbersome hardware. Simple obtainable process parameters together with the series neural network will contribute immensely in control and tuning of emissions for real-time vehicular applications.     

Mining churning behaviors and developing retention strategies based on a partial least squares (PLS) model

In a very competitive mobile telecommunication business environment, marketing managers need a business intelligence model that allows them to maintain an optimal (at least a near optimal) level of churners very effectively and efficiently while minimizing the costs throughout their marketing programs. As a first step toward optimal churn management program for marketing managers, this paper focuses on building an accurate and concise predictive model for the purpose of churn prediction utilizing a partial least squares (PLS)-based methodology on highly correlated data sets among variables. A preliminary experiment demonstrates that the presented model provides more accurate performance than traditional prediction models and identifies key variables to better understand churning behaviors. Further, a set of simple churn marketing programs—device management, overage management, and complaint management strategies—is presented and discussed.     

Improving customer acquisition models by incorporating spatial autocorrelation at different levels of granularity

Traditional CRM models often ignore the correlation that could exist among the purchasing behavior of surrounding prospects. Hence, a generalized linear autologistic regression model can be used to capture this interdependence and improve the predictive performance of the model. In particular, customer acquisition models can benefit from this. These models often suffer from a lack of data quality due to the limited amount of information available about potential new customers. Based on a customer acquisition model of a Japanese automobile brand, this study shows that the extra value resulting from incorporating neighborhood effects can vary significantly depending on the granularity level on which the neighborhoods are composed. A model based on a granularity level that is too coarse or too fine will incorporate too much or too little interdependence resulting in a less than optimal predictive improvement. Since neighborhood effects can have several sources (i.e. social influence, homophily and exogeneous shocks), this study suggests that the autocorrelation can be divided into several parts, each optimally measured at a different level of granularity. Therefore, a model is introduced that simultaneously incorporates multiple levels of granularity resulting in even more accurate predictions. Further, the effect of the sample size is examined. This shows that including spatial interdependence using finer levels of granularity is only useful when enough data is available to construct stable spatial lag effects. As a result, extending a spatial model with multiple granularity levels becomes increasingly valuable when the data sample becomes larger.     

Neural network based model predictive control for a steel pickling process

A multi-layer feedforward neural network model based predictive control scheme is developed for a multivariable nonlinear steel pickling process in this paper. In the acid baths three variables under controlled are the hydrochloric acid concentrations. The baths exhibit the normal features of an industrial system such as nonlinear dynamics and multi-effects among variables. In the modeling, multiple input, single-output recurrent neural network subsystem models are developed using input–output data sets obtaining from mathematical model simulation. The Levenberg–Marquardt algorithm is used to train the process models. In the control (MPC) algorithm, the feedforward neural network models are used to predict the state variables over a prediction horizon within the model predictive control algorithm for searching the optimal control actions via sequential quadratic programming. The proposed algorithm is tested for control of a steel pickling process in several cases in simulation such as for set point tracking, disturbance, model mismatch and presence of noise. The results for the neural network model predictive control (NNMPC) overall show better performance in the control of the system over the conventional PI controller in all cases.     

Applying Bayesian Belief Network approach to customer churn analysis: A case study on the telecom industry of Turkey

In telecommunication industry, for many organizations, it is really important to take place in the market. As competition increases between companies, customer churn becomes a great issue to deal with by the telecommunication providers. For an effective churn management, companies try to retain their existing customers, instead of acquiring new ones. Previous researches focus on predicting the customers with a propensity to churn in telecommunication industry. In this study, a model is constructed by Bayesian Belief Network to identify the behaviors of customers with a propensity to churn. The data used are collected from one of the telecommunication providers in Turkey. First, as only discrete variables are used in Bayesian Belief Networks, CHAID (Chi-squared Automatic Interaction Detector) algorithm is applied to discretize continuous variables. Then, a causal map as a base of Bayesian Belief Network is brought out via the results of correlation analysis, multicollinearity test and experts’ opinions. According to the results of Bayesian Belief Network, average minutes of calls, average billing amount, the frequency of calls to people from different providers and tariff type are the most important variables that explain customer churn. At the end of the study, three different scenarios that examine the characteristics of the churners are analyzed and promotions are suggested to reduce the churn rate.     

面向广义质量体系的软件质量贝叶斯评估模型

现有软件质量评估模型主要关注软件系统的基本质量特性,缺乏考虑顾客价值特性和开发商组织管理特性,不能全面和科学地评估软件系统质量.文章以贝叶斯网络刻画了广义质量特性变量间复杂依赖关系,构建了更有针对性的软件质量量化评估模型.应用实例表明,该模型能综合考虑软件系统广义质量特性,对软件质量作出合理评价,并能依据贝叶斯网络的反向推理功能找到影响软件质量的关键因素.     

Predicting subscriber dissatisfaction and improving retention inthe wireless telecommunications industry

We explore techniques from statistical machine learning to predict churn and, based on these predictions, to determine what incentives should be offered to subscribers to improve retention and maximize profitability to the carrier. The techniques include legit regression, decision trees, neural networks, and boosting. Our experiments are based on a database of nearly 47000 USA domestic subscribers and includes information about their usage, billing, credit, application, and complaint history. Our experiments show that under a wide variety of assumptions concerning the cost of intervention and the retention rate resulting from intervention, using predictive techniques to identify potential churners and offering incentives can yield significant savings to a carrier. We also show the importance of a data representation crafted by domain experts. Finally, we report on a real-world test of the techniques that validate our simulation experiments.     

Towards Circular Economy implementation: an agent-based simulation approach for business model changes

This paper introduces an agent-based approach to study customer behavior in terms of their acceptance of new business models in Circular Economy (CE) context. In a CE customers are perceived as integral part of the business and therefore customer acceptance of new business models becomes crucial as it determines the successful implementation of CE. However, tools or methods are missing to capture customer behavior to assess how customers will react if an organization introduces a new business model such as leasing or functional sales. The purpose of this research is to bring forward a quantitative analysis tool for identifying proper marketing and pricing strategies to obtain best fit demand behavior for the chosen new business model. This tool will support decision makers in determining the impact of introducing new (circular) business models. The model has been developed using an agent-based modeling approach which delivers results based on socio-demographic factors of a population and customers’ relative preferences of product attributes price, environmental friendliness and service-orientation. The implementation of the model has been tested using the practical business example of a washing machine. This research presents the first agent-based tool that can assess customer behavior and determine whether introduction of new business models will be accepted or not and how customer acceptance can be influenced to accelerate CE implementation. The tool integrates socio-demographic factors, product utility functions, social network structures and inter-agent communication in order to comprehensively describe behavior on individual customer level. In addition to the tool itself the results of this research indicates the need for systematic marketing strategies which emphasize CE value propositions in order to accelerate customer acceptance and shorten the transition time from linear to circular. Agent-based models are emphasized as highly capable to fill the gap between diffusion-based penetration of information and resulting behavior in the form of purchase decisions.     

A Comparison of Various Artificial Intelligence Methods in the Prediction of Bank Failures

The strong relationship between bank failure and economic growth attaches far more importance to the predictability of bank failures. Consequently, numerous statistical prediction models exist in the literature focusing on this particular subject. Besides, artificial intelligence techniques began to attain an increasing level of importance in the literature due to their predictive success. This study distinguishes itself from the similar ones in the sense that it presents a comparison of three different artificial intelligence methods, namely support vector machines (SVMs), radial basis function neural network (RBF-NN) and multilayer perceptrons (MLPs); in addition to subjecting the explanatory variables to principal component analysis (PCA). The extent of this study encompasses 37 privately owned commercial banks (17 failed, 20 non-failed) that were operating in Turkey for the period of 1997–2001. The main conclusions drawn from the study can be summarized as follows: (i) PCA does not appear to be an effective method with respect to the improvement of predictive power; (ii) SVMs and RBF demonstrated similar levels of predictive power; albeit SVMs was found to be the best model in terms of total predictive power; (iii) MLPs method stood out among the SVMs and RBF methods in a negative sense and exhibits the lowest predictive power.     

Data augmentation by predicting spending pleasure using commercially available external data

Since customer relationship management (CRM) plays an increasingly important role in a company’s marketing strategy, the database of the company can be considered as a valuable asset to compete with others. Consequently, companies constantly try to augment their database through data collection themselves, as well as through the acquisition of commercially available external data. Until now, little research has been done on the usefulness of these commercially available external databases for CRM. This study will present a methodology for such external data vendors based on random forests predictive modeling techniques to create commercial variables that solve the shortcomings of a classic transactional database. Eventually, we predicted spending pleasure variables, a composite measure of purchasing behavior and attitude, in 26 product categories for more than 3 million respondents. Enhancing a company’s transactional database with these variables can significantly improve the predictive performance of existing CRM models. This has been demonstrated in a case study with a magazine publisher for which prospects needed to be identified for new customer acquisition.     

The application of brute force logistic regression to corporate credit scoring models: Evidence from Serbian financial statements

In this paper a brute force logistic regression (LR) modeling approach is proposed and used to develop predictive credit scoring model for corporate entities. The modeling is based on 5 years of data from end-of-year financial statements of Serbian corporate entities, as well as, default event data. To the best of our knowledge, so far no relevant research about predictive power of financial ratios derived from Serbian financial statements has been published. This is also the first paper that generated 350 financial ratios to represent independent variables for 7590 corporate entities default predictions’. Many of derived financial ratios are new and were not discussed in literature before. Weight of evidence (WOE) method has been applied to transform and prepare financial ratios for brute force LR fitting simulations. Clustering method has been utilized to reduce long list of variables and to remove highly correlated financial ratios from partitioned training and validation datasets. The clustering results have revealed that number of variables can be reduced to short list of 24 financial ratios which are then analyzed in terms of default event predictive power. In this paper we propose the most predictive financial ratios from financial statements of Serbian corporate entities. The obtained short list of financial ratios has been used as a main input for brute force LR model simulations. According to literature, common practice to select variables in final model is to run stepwise, forward or backward LR. However, this research has been conducted in a way that the brute force LR simulations have to obtain all possible combinations of models that comprise of 5–14 independent variables from the short list of 24 financial ratios. The total number of simulated resulting LR models is around 14 million. Each model has been fitted through extensive and time consuming brute force LR simulations using SAS® code written by the authors. The total number of 342,016 simulated models (“well-founded” models) has satisfied the established credit scoring model validity conditions. The well-founded models have been ranked according to GINI performance on validation dataset. After all well-founded models have been ranked, the model with highest predictive power and consisting of 8 financial ratios has been selected and analyzed in terms of receiver-operating characteristic curve (ROC), GINI, AIC, SC, LR fitting statistics and correlation coefficients. The financial ratio constituents of that model have been discussed and benchmarked with several models from relevant literature.