M.S. Applied Data Science - Capstone Chronicles 2025

35

Figure 19 OOB Error VS. Number of Trees Across Feature Strategies

5.4 Final Model’s Feature Importance Analysis Feature importance was evaluated to identify the most influential variables contributing to the final model’s predictions. Depending on the model type, importance was derived either from built-in attributes such as feature_importances_ for tree-based models or from absolute coefficient values for linear models like logistic regression. If feature selection was applied, only the selected features were considered in the analysis. In cases where neither method was available, selection frequency during cross-validation was used to

assess relevance. This ensured that a meaningful importance ranking was available regardless of the modeling approach. The top features included reason_word_count, month_sin, and month_cos, suggesting that both textual and temporal variables played key roles in the model’s performance. Figure 20 demonstrates the final rankings, visualized using a bar chart, which was saved for documentation. This visualization provides a clear summary of the top 20 features and supports interpretation of the model’s decision-making process.

39