Interpretable Machine Learning in Healthcare

In this research, we aimed to explore the importance of interpretability and explainability in machine learning models and how these concepts are especially crucial in the healthcare sector. While high accuracy is important in all industries, in healthcare, it is essential to understand the reasons behind a prediction to make informed decisions that affect human lives.

 

To address this issue, we utilized interpretable machine learning (IML) methods, such as SHAP and Skater, to extract the interpretation for each classification model in our study. Our goal was to make the models more understandable, enabling physicians to diagnose cervical cancer accurately. We focused on cervical cancer diagnosis due to its significant impact on women's lives, with it being the fourth leading cause of death among women globally.

 

Our study found that the decision tree model performed better than the ensemble model, particularly the Voting Classifier. This result highlights the importance of choosing a model that suits the problem at hand, rather than relying on ensemble models in all scenarios. Additionally, our interpretation results showed that while IML methods can identify important features, they may not always capture the true causal relationships between features and target variables. This finding emphasizes the need for more robust causal inference algorithms and the integration of expert knowledge to ensure correct causal inference.

 

In conclusion, our study highlights the importance of interpretability and explainability in machine learning models in healthcare, specifically for diagnosing cervical cancer. We recommend considering the limitations of existing IML methods and the need for more advanced causal inference algorithms and expert knowledge integration to derive accurate causal relationships. Overall, our study contributes to the ongoing efforts to develop a new digital screening solution for cervical cancer, making screening more accessible to women in areas with scarce healthcare resources.