The rapid growth of mobile internet and social media usage in Sub-Saharan Africa has amplified the need for accurate Quality of Experience (QoE) assessment in resource-constrained network environments. This paper introduces a novel hybrid Machine Learning (ML) framework for predicting user QoE in Cameroon’s Third Generation and Fourth Generation (3G/4G) networks, leveraging a unique real-world dataset that integrates crowdsourced Quality of Service (QoS) measurements with subjective user satisfaction surveys. Addressing the limitations of existing QoE studies that focus on well-resourced regions, our approach proposes QoE Predictor (QoEPredict). This stacking ensemble combines eXtreme Gradient Boosting (XGBoost) and Random Forest classifiers with an XGBoost meta-learner. A key innovation is the use of disagreement features to capture divergences between base model predictions, allowing the meta-learner to resolve conflicts and enhance predictive accuracy. The proposed five-stage pipeline incorporates data preprocessing, feature engineering via Uniform Manifold Approximation and Projection (UMAP), unsupervised clustering, and Bayesian hyperparameter optimisation using Hyperopt, ensuring a robust and transferable methodology. Explainable AI (XAI) is integrated through SHapley Additive exPlanations (SHAP) analysis to provide feature-level interpretability and actionable insights for network operators. An experimental evaluation of 1,934 user sessions demonstrates that QoEPredict achieves a 90% F1 Score and accuracy, outperforming single-model baselines across all metrics. This work represents one of the first large-scale, interpretable QoE prediction frameworks for mobile social media applications in Sub-Saharan Africa. By combining ensemble modelling with explainability and contextualised insights, the study offers both methodological advances and practical guidance for implementing QoE-aware network management strategies in developing regions facing infrastructural and operational constraints.