Authors: Badisa. Adhilakshmi, Mula Srilatha, Golla Manusri, Bodepudi Tejaswini, Daggubati Maneesha
Abstract: Faking resumes is one of the greatest challenges in the contemporary recruitment systems where most applicants tend to embellish or lie about their academic and professional experience or technical abilities in order to have an advantage in employment. The manual verification systems are tedious, time consuming and they are also subject to error, which makes them ineffective in large-scale hiring. Previous automated systems based on classical machine learning systems like Support Vector Machines (SVM) or Random Forest are only capable of dealing with structured data and do not effectively deal with unstructured, multilingual and complex resumes. The consequences of these limitations are low accuracy, low contextual knowledge and low scalability. To address these issues, in this paper, a hybrid AI-inspired resume verification system incorporating the methods of Natural Language Processing (NLP), deep learning, and classical machine learning will be suggested. The system preprocesses resumes of different types (PDF, DOCX, text) and finds significant data, including education, skills, and experience, and it describes it with contextual embeddings with Transformer-based models. Convolutional Neural Networks (CNNs) are used to capture local linguistic patterns whereas traditional ML models like Random Forest and Gradient Boosting are used to analyse engineered numerical features. An ensemble classifier is a stacked ensemble of these components that is used to give a final score of authenticity, or what percentage probability a resume is a fake resume. The experimental evidence shows that the hybrid model is much better in comparison to traditional methods, as the accuracy of the models is 85-95 with the greatest accuracy of the Transformer based model of 94, and better precision, recall, and F1-score. High-performance, scalable, and automated approach to resume fraud detection through the combination of NLP, deep learning, and classical ML will make recruitment processes more efficient, transparent, and more credible.
DOI: https://doi.org/10.5281/zenodo.20066837
