Customer Churn Prediction in FMCG Sector Using Machine Learning Applications

Abstract

Non-contractual setting and many brands and alternative products make customer retention relatively more difficult in the FMCG market. Besides, there is no absolute customer loyalty, as most buyers split their purchases among several almost equivalent brands. Thereby, this study aims to probe the contribution of various machine learning algorithms to predict churn behaviour of the most valuable part of the existing customers of some FMCG brands (detergent, fabric conditioner, shampoo and carbonated soft drink) based on a real dataset obtained in the Turkish market over the two successive years (2018 and 2019). In this context, exploratory data analysis and feature engineering are carried out mostly to build many predictive models to reach consistent and viable results. Further, RFM analysis and clustering techniques with K-Means clustering are employed to generate meaningful insights for business operations and marketing campaigns. Lastly, revenue contributions of improved customer retention can be achieved, utilising actionable intelligence created by the churn prediction.

A data science-based marketing decision support system for brand management

“…The customer classification agent classifies methods based on the consumer database in order to identify the main characteristics of customers who, depending on the purpose of the classification, have, for example, a higher risk of refusing the enterprise's services (the risk of churn) or a different probability of a positive reaction to advertising mailings. This agent can identify the most valuable customer groups (Günesen et al, 2021) and the further development of appropriate marketing measures (Mo & Yang, 2022). Like customer segmentation, classification methods should also be applied regularly to maintain relevance to the market situation.…”

Customer churn prediction model: a case of the telecommunication market

“…Tianyuan et al (2021) proved that the most widely used data mining methods for predicting customer outflows are logistic regression, decision tree, support vector machine (SVM). In the last decade, many scientists present the application results of various machine learning methods and algorithms for classification (Bandam et al, 2022) and prediction of churn behavior of the most valuable part of the current clients (Günesen et al, 2021), searching for more efficient approaches of customer churn prediction. Among the used methods and models for solving the tasks of data processing for telecommunication industry are logistic regression, decision tree and random forest models for churn prediction (Kiguchi et al, 2022;Vezzoli et al, 2020;Kuznietsova et al, 2022), K-means, SVM (Sánchez et al, 2022), the combination of k-means customer segmentation and SVM prediction (Xiahou et al, 2022), the multi-level classification using SVM in the SLS-SVM algorithm (Huang, 2022), the Naïve Bayes for prediction of loyal or disloyal customers and their behavior (Jayadi et al, 2020;Rabiul Alam et al, 2021), an ECHAID (exhaustive Chi-square automatic interaction detector) classification tree for consumers segmentation (Kelley et al, 2022) and so on.…”

“…In order to solve the low stability and efficiency, high complexity of marketing decisions, a marketing adaptive decision-making algorithm based on big data analysis is proposed by Lv (2022). Using intelligence results from customer churn prediction models, there is the significant potential to generate additional revenue by improving customer retention strategy (Günesen et al, 2021). Churn prediction model will help telecom enterprises to effectively forecast the opportunity of customer's churn, take appropriate targeted measures and make better marketing decisions to avoid the outflow of customers and, as a result, increase their profits.…”