<p>Social media has become a significant part of people’s lives, leading to the rapid spread of false information, such as rumors, which negatively impact society and individuals. Therefore, it is crucial to detect such rumors at an early stage. This study proposes a novel approach for explainable rumor detection by integrating topic modeling with Local Interpretable Model-agnostic Explanations (LIME). Our approach employs an unsupervised machine learning technique, specifically Latent Dirichlet Allocation (LDA), to uncover hidden topics within rumor data. These topics serve as features for classification. To ensure stability, we utilize a Random Forest classifier with 5-fold cross-validation, achieving a superior accuracy of 93.25% on the PHEME dataset compared to other state-of-the-art models. The combination of topic-based classification enhances the accuracy and interpretability of rumor detection model. Additionally, our model offers greater interpretability than traditional LIME methods. While LIME provides local explanations that may vary for each instance, our method captures stable topic distributions that are particularly effective for early-stage rumor detection with better explanations.</p>