Smartphone Brand Loyalty Prediction Using Behavioural and Preference Data

Abstract

Knowing ahead of time whether a student will stick with their current smartphone brand or switch at their next upgrade is something marketers genuinely want to predict. This paper tackles exactly that question. We put together a full machine learning pipeline, ran it on survey data from 2,000 student respondents, and went head-to-head with four classifiers: Logistic Regression, Random Forest, Gradient Boosting, and Extra Trees. The pipeline was built with one particular concern in mind data leakage which is a surprisingly common mistake in survey-based classification studies and tends to produce accuracy numbers that look great on paper but fall apart in practice. Eight behavioural and attitudinal features served as the model inputs, each encoded according to its type. Gradient Boosting won the comparison, reaching 90% test accuracy with an F1-score of 0.9180 and a ROC-AUC of 0.9535. Worth noting: hyperparameter tuning actually made things slightly worse, which tells us that the defaults were already near the sweet spot for this particular dataset. A Streamlit web interface was also developed so that people who do not write code can still run predictions and explore results on their own.