Figure 2
The diagram illustrates a data processing framework starting with qualitative data sources such as Facebook group posts and comments, and student membership counts. These sources are encoded using indicator coding and then aggregated weekly. The aggregated data is used as input for three models: Persistence Baseline, Random Forest, and MLP. The outputs of these models are evaluated using time-aware evaluation metrics such as MAE, RMSE, and R squared. The results are further analyzed using permutation importance and sensitivity analysis.End-to-end pipeline of the proposed framework