In real-world datasets, class imbalance is common, where certain classes are underrepresented, leading to skewed distributions that negatively impact classifier performance, particularly for minority classes. This issue is prevalent in script identification tasks, where underrepresented scripts lead to biased models that struggle to predict minority classes accurately. To address this problem, we explored the effectiveness of various resampling techniques, grouped into under-sampling, over-sampling, and hybrid-sampling methods. Our study evaluates these techniques by testing multiple classifiers on a subset of the MDIW-13 dataset, focusing on handwritten page level script identification. The results demonstrate significant improvements in various performance metrics when applying resampling techniques, emphasizing the crucial role of hybrid sampling in mitigating class imbalance in script identification tasks.