A Hybrid Ensemble Learning Framework for Semantic Classification of Ancient Indian Cosmological Texts using SVM–XGBoost Fusion

Abstract

The convergence of artificial intelligence and traditional knowledge systems opens up new avenues for cross-disciplinary research. The present work describes a hybrid ensemble learning-based framework for semantic classification of domain-specific texts (derived from a classical Indian astronomical treatise सूर्य सिद्धांत. The main aim is to convert unorganized ancient textual knowledge into structured, machine-readable representations by applying effective and scalable machine learning methodologies. The present method is based Term Frequency–Inverse Document Frequency (TF-IDF) for feature representation and two machine learning models: Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost). To improve classification results, a hybrid ensemble approach is adopted, aggregating the probabilistic predictions of the two models in a weighted way. This integrative process exploits the linear decision function of SVM and the non-linear learning perturbation of XGBoost to enhance stability and predictive performance. The dataset is carefully selected and labelled with a variety of semantic categories, such as astronomy, cosmology, mathematics, time, history and general information. Evaluation is performed with the usual metrics of accuracy, precision, recall and F1-score. Results show that the ensemble model leads to the best classification results and that it is computationally efficient. Our framework enables a scalable and interpretable solution for domain-specific text classification for low-resource, culturally-rich data. In contrast to transformer-based methods with high computational cost, the proposed method achieves competitive performance and a better trade-off between efficiency and effectiveness. This work advances the research areas of natural language processing, computational humanities, and cultural knowledge preservation by facilitating the intelligent analysis of ancient scientific texts. The proposed SVM–XGBoost ensemble model attained an accuracy of 98% with a macro F1-score of 0.92 and weighted F1-score of 0.98, which indicates the robustness and better performance than the individual based models.