AAI_2025_Capstone_Chronicles_Combined

Cinema Analytics and Prediction System

29

Revenue Prediction and Success Classification

For the regression task, the quality of predictions is represented in Figure 28, where

actual vs. predicted revenues are plotted alongside the

budget. The model captured overall revenue trends but

struggled with extreme cases, underestimating

blockbusters and overestimating some low-revenue

films due to their rarity in the training data. Still, it

Figure 28: Actual vs Predicted revenue

outperformed mean and median baselines across all

error metrics (Figure 29) like MAE, MedAE, and MAPE. Its relative errors (Figure 30) were much

lower, and it achieved an R-squared score near 0.7, showing strong performance compared to

the negative R-squared of baseline models .

Figure 30: Error metrics comparison

Figure 29: Error metrics comparison

The binary XGBoost model for success classification performed well, with balanced

precision, recall, and F1-scores around 0.75, showing good generalization. In contrast, the

multiclass XGBoost model showed mixed performance. The model's predictions varied

significantly across categories. It performed reasonably well on the Flop and Superhit classes,

achieving F1-scores of 0.78 and 0.65, respectively. However, it struggled to classify the Hit

category, with an F1-score of just 0.09, suggesting potential issues such as class imbalance or

ambiguous feature representations for that class.

196

Internal