Haematological Whole Slide Images present a critical challenge due to severe class imbalances and significant variability within each class, which can lead to biased models and inaccurate diagnoses. To address these challenges, this research will explore advanced class balancing techniques, such as over-sampling and cost-sensitive learning, alongside novel strategies like instance-aware sampling. Additionally, it will integrate attention mechanisms within ensemble models to enhance the model’s ability to generalise across diverse cases, ultimately aiming to improve diagnostic accuracy in haematology.