International Research Journal of Education and Technology

This article discusses how to transform a black box into a more interpretable model than a white box, especially in the field of bioinformatics, using artificial intelligence (XAI). Black box models such as deep learning and integration are quite accurate but lack transparency, which prevents them from being used in important applications such as healthcare. In our previous article, we proposed XAI techniques that can transform these opaque models into more understandable models, allowing researchers to understand decision-making models. In this work, we apply XAI techniques to black box and white box models in bioinformatics, focusing on accurate measurement and interpretation. We use SHAP values and visual models to highlight the impact of various aspects of the prediction model by experimenting with random forests (as a black box model) and logistic regression and decision trees (as a white box model). Our results show a trade-off: models with black bars (such as random forests) generally achieve higher accuracy in capturing patterns of complex objects, while models with white lines provide a clear ordering process that is crucial for clear understanding. This paper highlights the utility of XAI and highlights the importance of translation in bioinformatics to ensure that AI-driven models are not only accurate, but also interpretable and reliable for these important applications.

KEYWORDS: Explainable AI (XAI), Logistic Regression, Decision tree, White Box Model (WBM), Black Box Model (BBM)