Blood Disease Risk Assessment – a ComparativeAnalysis of Machine Learning Models and XAIbased Model Interpretability

Healthcare sector has always been one of themost important industries for humankind. As ArtificialIntelligence is automating several industries, use of AI andmachine learning based algorithms in healthcare is anincredible idea to improvise the way we look towards andperform healthcare. One of the best ways to bring an AI/MLbased change in healthcare is through first-hand diagnosis ofdiseases. There are numerous blood related diseases, presence,or risk of which can be assessed through analysing the basicblood report of a patient. Developing a machine learningmodel to perform this analysis and flash the comments ‘risk’or ‘no risk’ on screen in seconds would be a big advancementin the diagnosis sector, saving a lot of time and man force. Thisresearch paper develops machine learning models to performthis first-hand blood related disease risk assessment and alsoexplores the processing of these black-box models throughexplainable AI (XAI) to ensure the trustworthiness of thesemodels for using them at scale. The machine learning modelstrained and tested for the blood disease risk assessment areLogistic Regression, Decision Tree, Random Forest andXGBoost and the best test accuracy obtained from thesemachine learning models is 100%. The feature contributingthe most to the prediction as recorded by XAI are MCV,MCH, HBG and HBA.