A Machine Learning Framework for Predicting Customer Purchase Intention on Jumia Nigeria Using Customer Review Analytics

Abstract

Customer reviews have become one of the richest sources of behavioural signal available to online retailers, yet many e-commerce studies in the Nigerian context continue to rely on small-sample survey instruments to explain purchase-related outcomes. This study departs from the survey tradition and instead develops a machine learning framework that predicts customer purchase intention directly from unstructured customer review text collected from a Jumia online marketplace. A corpus of 3,801 cleaned product reviews, each paired with a customer-assigned star rating, was used to construct a binary purchase-intention label (positive when the rating was four or five stars, negative otherwise). Review headlines and bodies were cleaned, tokenised and transformed into Term Frequency-Inverse Document Frequency (TF-IDF) vectors with unigram and bigram features. Five supervised classifiers: Logistic Regression, Multinomial Naive Bayes, Linear Support Vector Machine, Random Forest, and Gradient Boosting were trained and compared on an 80/20 stratified train-test split using accuracy, macro-averaged precision, recall, F1-score, and weighted F1-score. Logistic Regression achieved the strongest balance of performance (accuracy = 0.862, macro F1 = 0.794), followed closely by the Linear Support Vector Machine, which achieved the highest raw accuracy (0.882) but a lower macro F1 due to the class imbalance in the data. Feature-importance analysis revealed that terms expressing product quality, delivery satisfaction, and value-for-money were the strongest linguistic predictors of positive purchase intention, while terms referencing damage, delay, and counterfeit concerns dominated the negative class. The findings demonstrate that review-analytics-based machine learning pipelines offer a scalable, low-cost alternative to survey-based approaches for monitoring purchase intention on African e-commerce platforms, with clear implications for platform-level recommendation, quality-control, and customer-retention strategies.