Classification of Imbalanced Immunotherapy and Health-Related Data Utilising Novel Machine Learning Experiments

Machine learning is applied in healthcare to enhance the diagnosis of immunotherapy treatments. Precise classification of imbalanced immunotherapy data is essential, as the performance of the diagnosis results is affected if a class dominates. However, the classification of imbalanced datasets is challenging, as only implementing algorithms is insufficient as the results are less optimal. This study aimed to design, conduct and evaluate experiments, addressing the impacts of imbalanced data on classification. Five novel machine learning experiments are performed observing imbalanced datasets of immunotherapy, diabetes, cryotherapy and ’one imbalanced dataset’ utilising Weka and Python. The experiments are: reproducing a study from published literature, class balancing, comparing and integrating imbalanced datasets, and exploring the effect of minority class on classification. Experiments show that specificity and sensitivity are essential for imbalanced datasets as accuracy alone is less reliable. By applying SMOTE and oversampling specificity and sensitivity are improved.