Revisiting SVMs in Detecting Lies: (In)Validating Two Models

The task in this paper is to re-evaluate Burgoon’s (2012) model and a proposed linguistic model to lie detecting (see El-Zawawy, 2017) and to test their validity with the aid of one well-documented machine learning method, namely Support Vector Machine (henceforth SVM). The two corpora chosen are false statements by Joseph Biden and Donald Trump, all delivered spontaneously, to see how the proposed model as aided by SVM can detect their falsehood. Results show that the Burgoon model demonstrates stable and reliable performance, achieving high accuracy, precision, and recall, particularly in holistic analysis, where it records 91.84% accuracy and an F1-score of 90.91%. In contrast, the Proposed Model shows extreme inconsistencies, with an overall accuracy of 0.00%, suggesting that its predictions are dubious