Comparing Best Subset and Lasso Regression in the Customer Loyalty Prediction in a Restaurant Dataset

Abstract

Independent variables such as attributes related to the product, service quality, and purchase satisfaction are often correlated with one another in a customer loyalty research case. For instance, product attributes may overlap with service quality, and both factors jointly influence purchase decisions. To address multicollinearity, models such as best subset and Lasso regression can be employed. These models will be applied to a restaurant customer loyalty dataset. This study was conducted at Warung Tuman Restaurant in South Tangerang, Indonesia, from April to June 2022. We analyze responses from 100 purposively sampled consumers, with loyalty as the dependent variable and  (product attributes),  (service quality), and  (purchase satisfaction) as predictors. Correlation analysis revealed strong positive relationships (r = 0.44, p < 0.00) among predictors, confirming multicollinearity and justifying the use of best-subset and Lasso. The dataset was split into a 60% training set and a 40% test set, with the training set used to develop predictive models, which were then evaluated for accuracy using the test set. All correlation values demonstrate a significant positive relationship between the independent variables, indicating the suitability of the best subset and Lasso regression applications. The best subset and Lasso regression generate models with two independent predictor variables, i.e. product attributes and purchase satisfaction. The best subset regression exhibits a lower Sum of Squared Errors (SSE), thereby indicating its superior performance compared to the Lasso regression model. To effectively sustain and improve customer loyalty, restaurant managers should prioritize optimizing product attributes and purchase satisfaction factors.