Confusion Matrix

Due to the unbalanced nature of the data lower class label count were over sampled in a separate model.

The oversampling resulted in fewer correct predictions for class label 5, but the model showed better percentage results for class 4 by a significant amount.

Metrics

• Mean Absolute Error (lower is better)
• Mean Zero-One Error (lower is better)
• Precision calculates the proportion of positive predictions, those are actually correct
• Recall calculates the proportion of positive predictions, those are identified correct
• F1 Score is a good matrix for imbalanced data as it works in favor of class label with lower count