Handling Imbalanced Classification Datasets

Accuracy is misleading because it is computationally slower than Mean Squared Error on large arrays.

Accuracy can reach 99.9% by simply predicting 'No Fraud' for every single transaction. Precision, Recall, and F1-Score are preferred metrics.

Accuracy scales with batch sizes, so model compilers overfit the neural weights. R-Squared is preferred.

Accuracy ignores categorical string parameters, which throws TypeErrors. Gini Impurity is preferred.