The integrity of modern electronic systems is increasingly being endangered by malignant mutations which are commonly referred to as Hardware Trojans (HTs). Detection of such threats is a complex challenge due to the stealthy nature of Trojans, the diversity of circuit topologies and the opacity of Machine Learning (ML) classifiers. This paper presents an explainable Machine Learning-based framework that incorporates three complementary categories of features such as structural features, Graph Centrality Measures (GCMs) extracted from directed graph transformations and Functional features from gate-level netlist to detect Trojans. The proposed methodology evaluates the combined influence of each feature category on detection accuracy and interpretability, unlike existing approaches that analyze them in isolation or consider only a subset of these feature groups. The eXtreme Gradient Boosting (XGBoost) model is employed as the classifier, and three feature ranking strategies such as SHapley Additive explanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and XGBoost’s intrinsic importance are utilized for their ability to rank features there by indicating the most influential ones and provide transparent insights into the ML model’s behaviour as they are black-box in nature. Filter-based ranking-driven feature selection technique is used to further eliminate redundancy. Beyond conventional evaluation metrics, the study also evaluates feature ranking effectiveness using fidelity, simplicity, stability and coverage, collectively known as eXplainable AI (XAI) metrics marking the first application of such metrics in Trojan detection. Experimental validation on sequential Trust-HUB benchmark circuits demonstrates the efficacy of the proposed approach that establishes a principled path towards explainable and reliable HT detection.