Understanding and predicting user engagement and behaviour requires sentiment analysis. The main objectives of this study were to enhance the Random Forest (RF) algorithm for sentiment prediction in social media engagement, apply synthetic minority oversampling technique (SMOTE) to address class imbalance in the dataset, and utilise SHapley Additive ExPlanations (SHAP) for model interpretability. Unlike conventional methods targeted at text data, this research incorporates engagement metrics and demography for improved results in prediction accuracy and additional knowledge. This research differs from the existing models, both in its multi-faceted approach, combining RF, demographic metrics, and explainable artificial intelligence (XAI) to boost interpretability and address the limitations of previous models. Our results outperform existing studies in accuracy and provide actionable explanations behind decision-making.