Application Of Machine Learning Algorithms in Predicting Customer Loyalty Towards Grocery Retailers

Abstract

Retailers strive for customer loyalty in the sense of repeat purchases, but also as a high proportion of purchases (compared to competitors) and willingness to recommend to other customers. This paper examines customer loyalty in the grocery sector as a three-dimensional construct and shows how machine learning techniques can be useful in its study. Price characteristics of the retailer (price level, value for money, price dynamics, price communication and price dispersion) and non-price characteristics of the retailer (general product range, retailer's private label product range, store design and atmosphere, service level and location) are included in the model as predictor variables. Using the data collected through the primary research conducted in Croatia, 433 samples were divided into 10 independent predictor variables and one dependent variable (customer loyalty), a prediction was created using supervised machine learning classification algorithms. The Random Forest classifier proves to be the best choice overall, with ROC_AUC value of 0.790, a high accuracy of 0.915 and an F1 score of 0.954, reflecting both precision and responsiveness. The application of the SHapley Additive exPlanations analysis additionally enables the interpretation of the results, highlighting the influence of features on the accuracy of the prediction. The results indicate that price dynamics and service level are the most important features for the model predictions, followed by value for money and price communication.