<p>In panic-buying situations, individuals suddenly purchase excessive quantities of goods, leading to a massive crisis of essential goods in the market. As a result, many consumers cannot access the required products, creating an unstable societal situation. Despite the importance of this issue, only limited research has focused on providing automated solutions for detecting panic-buying behavior. This work proposes a machine learning-based model to predict panic-buying behavior, evaluate the outcomes of classifiers, interpret the classification results, and identify relevant factors for this situation. In this work, we collected customer purchasing records of COVID-19 from a public repository^{<CitationRef CitationID="CR1">1</CitationRef>}. This primary dataset was preprocessed and generated several SMOTE variants. Multiple feature selection methods were employed on these balanced datasets to create feature subsets. A range of state-of-the-art classifiers was then applied to each balanced dataset, both with and without fine-tuning, to explore the best classification algorithm. The post-hoc statistical significance test, the Friedman test, was conducted to validate classifiers performance. Furthermore, an explainable AI method was employed to interpret the top-performing models and uncover the key factors to generate these results. After observing the outcomes of classifiers for each balanced variant, Gradient Boosting and its advanced variants consistently outperform other models, demonstrating robustness and stability in detecting panic-buying behaviors.</p>