Table 12 Potential risks of overfitting and mitigation Strategies.

Single dataset usage (BR35H)

Model trained and tested on one dataset may not generalize to other data distributions.

Reduced generalization and possible dataset bias.

Future validation on additional datasets (e.g., Figshare, BraTS).

Limited label diversity (binary classification)

Dataset only includes tumor vs. non-tumor categories, lacking multi-class complexity.

Model may not perform well on multi-type tumor detection tasks.

Extend framework to multi-class or multi-modal MRI datasets.

High model capacity

Deep models can memorize patterns instead of learning generalized features.

Overfitting to training data, inflated accuracy.

Applied transfer learning, layer freezing, and pruning to reduce overfitting.

Small sample size

Smaller datasets increase variance and sensitivity to noise.

Unstable model performance across samples.

Use of data augmentation and repeated runs to improve reliability.