TOWARDS EXPLAINABLE AND INFERENTIAL BRAND-AWARE INTELLIGENT MODELS FOR PREDICTING CUSTOMER CONSUMPTION BEHAVIOR

ABSTRACT: Customer consumption behavior is a key factor in understanding and strengthening the relationship between customers and companies. Among the behaviors that pose a threat to businesses is customer churn. Being able to predict churn behavior in advance helps companies identify potential leavers early, allowing them to take proactive, data-driven actions to reduce customer loss, improve loyalty, and minimize costs and revenue losses. This study aims to develop explainable and inferential brand-aware intelligent models capable of predicting customer consumption behavior by classifying customers according to whether they will churn or remain loyal. The research applies a descriptive and analytical approach, using a real behavioral dataset from a telecommunications company. Four intelligent models were evaluated: Decision Tree (DT), Gradient Boosting Machine (GBM), Logistic Regression (LR), and Random Forest (RF). These models were compared using key performance metrics, explainability metrics, and hypothesis testing to determine the most effective and statistically significant predictive model capable of providing interpretable insights into customer behavior. Explainability techniques, including Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), revealed that Contract Type, Monthly Charge, and Tenure in Months were the most influential factors driving churn decisions, while local interpretability confirmed individual-level behavior patterns consistent with global trends. Hypothesis testing further validated the significance of these variables (p < 0.001), confirming strong statistical relationships between key behavioral attributes and churn outcomes. The results indicate that GBM achieved the highest accuracy rate of 83% with a standard deviation of 0.01, confirming both the stability of the data and the robustness of the model. The integration of explainability and inferential analysis provides deeper insights into customer decision dynamics and supports more transparent, evidence-based marketing strategies.