Authorship Classification in Academic and Scientific Documents: A Machine Learning-Based Approach

This paper presents a machine learn-ing-based system that incorporates text mining to analyze and classify writing styles in scientific reports authored by fac-ulty members at the Pontificia Universidad Católica del Ecuador, Esmeraldas. The sys-tem aims to enhance academic integrity by identifying potential cases of false author-ship. A dataset of research papers written in Spanish by faculty professors was pro-cessed using TF-IDF (Term Frequency-In-verse Document Frequency) and Word Em-beddings for feature extraction. To assess classification performance, seven super-vised learning models were tested: Linear Support Vector Classifier (SVC), SVC with RBF kernel, Random Forest, Decision Tree, Logistic Regression, k-Nearest Neighbors (k-NN), and Naïve Bayes. The Logistic Re-gression model yielded the highest accura-cy (89.62%), closely followed by Linear SVC (87.36%) and RBF SVC (86.59%), outper-forming tree-based and probabilistic meth-ods with statistical significance (p