Predicting Consumer Purchase Intention in Informal Retail Using Machine Learning and the Purchase Intention Probability Index (PIPI)

Abstract

Informal retail remains a growing and predominant form of shopping for many people. However, modern and well-organized supermarkets, using data-driven approaches to attract consumers, have increasingly challenged informal retailers in recent years. This phenomenon presents new challenges, particularly in predicting consumers' purchase intentions given limited, unstructured, and poorly documented data. Therefore, this study aims to develop and evaluate a predictive model for consumer purchase intention in informal retail using machine learning techniques and to introduce the Purchase Intention Probability Index (PIPI) as a probability-based aggregation approach to enhance predictive sensitivity. The study uses the Subsistence Retail Consumer Dataset from Mendeley Data, comprising 281 consumer records with 38 demographic, behavioral, and psychological attributes, with purchase intention as the binary target variable. Three widely used classification algorithms in consumer behavior research (decision tree, random forest, and support vector machine (SVM)) were employed to identify purchase-predictive patterns in the data. Based on these models, the PIPI was developed, which aggregates the highest probabilities from all three models to produce more robust predictions, particularly for small and heterogeneous datasets, and supports cross-model performance evaluation. The results show that the proposed PIPI method achieves the highest recall (1.00), outperforming individual classifiers in detecting purchase intention. This fact indicates that informal retailers can apply machine-learning-based analytics to improve marketing effectiveness and decision-making without requiring advanced technological infrastructure.