In the medical world, stroke is one of the most prominent diseases that threaten human life in the world. The most common causes of death in the world are stroke, heart disease and cancer. Fortunately, stroke can be prevented if high risks patients are identified before the occurrence of a stroke. A medical diagnosis for stroke prediction is a complex process. However, machine learning algorithms can be used to facilitate stroke prediction in the early stages. The performance of the classification algorithms depends on a balanced dataset hence this study aims to implement the imbalanced data techniques as undersampling, oversampling and hybrid techniques to enhance the classification models performance.