Interpretability of Machine Learning Algorithms for News Category Classification Using XAI

Text classification has been a popular research topic for several years. To date, several advanced models have been developed in this field. In today's digital landscape, the proliferation of online news sources necessitates the efficient categorization of user accessibility. To address this situation, machine learning models can be used to automate news category classification based on news headlines and short descriptions. However, we know that machine-learning models act like black boxes. The interpretability of a model provides a clear understanding of how decisions are made, ensuring transparency, and user acceptance. This transparency helps in model validation and effective collaboration between human and computer interactions. Local Interpretable Model-agnostic Explanations (LIME), as an Explainable Artificial Intelligence (XAI) technique, can be implemented to generate interpretable explanations that aid in understanding the reasoning behind specific predictions. Our approach aims not only to predict news categories but also to provide transparent insights into model decisions. Initially, we used Sentence Bidirectional Encoder Representations from Transformers (SBERT) for contextual text embedding purposes, followed by different machine learning models for classification tasks, such as Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), and K-Nearest Neighbors (KNN). Notably, RF achieves exceptional accuracy of 91.48 %, surpassing contemporary benchmarks. Finally, LIME elucidates crucial features guiding classification decisions, facilitating model validation, and fostering human-computer collaboration. This study enriches the evolving discourse on interpretable AI models by providing a robust framework for transparent news classification in an era inundated by information.