Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{202776,
        author = {Prasad Vaibhav Pande and Prof. Vivekanand Thakare and Aman Sunil Bodkhe and Paras Ranjit Patil and Santoshi Vilas Mendhe},
        title = {AI-DRIVEN SUBJECTIVE ANSWER EVALUATION SYSTEM},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {7360-7369},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=202776},
        abstract = {Descriptive answer evaluation in academic settings has long remained a manual, time-consuming, and subjectively inconsistent process. As student enrolments scale globally, educators face mounting pressure to deliver timely, unbiased, and constructive assessments a challenge that existing cloud-dependent AI grading solutions fail to resolve without incurring significant data privacy risks and recurring financial costs. This paper introduces a locally deployable, fully offline AI-driven framework for automated subjective answer evaluation, engineered entirely upon open-source Natural Language Processing (NLP) and lightweight Machine Learning (ML) libraries. The proposed system assesses student responses through six independently weighted evaluation dimensions, namely Semantic Similarity, Keyword Coverage, Relevance Detection, Grammar Analysis, Writing Quality, and Originality, whose scores are aggregated via a configurable weighted linear formula to derive a final grade mapped across an A+ to F scale. Semantic alignment between student and model answers is computed through TF-IDF vectorization paired with Cosine Similarity, while domain-specific keyword presence is verified using spaCy-based Named Entity Recognition combined with lemmatization. Linguistic quality is assessed through Language Tool grammar analysis, and verbatim copying is flagged through N-gram overlap and Jaccard Similarity metrics. The system further incorporates a dual-engine Optical Character Recognition pipeline using Tesseract and EasyOCR to support handwritten and scanned answer sheet processing. Experimental validation on a dataset of 150 student responses across multiple academic disciplines yielded a Pearson correlation coefficient of 0.83 against expert human-assigned grades, with a Mean Absolute Error of 6.2 percentage points. A pilot user study confirmed that 85% of educators found the system's output consistent with their own grading, while 91% of students rated the dimension-specific AI feedback as helpful for identifying knowledge gaps. The system demonstrates that accurate, explainable, and ethically responsible automated grading is achievable without cloud infrastructure, external API dependency, or recurring cost.},
        keywords = {Automated Subjective Evaluation, Natural Language Processing, TF-IDF Vectorization, Cosine Similarity, Offline Grading System, Optical Character Recognition, Explainable AI, Educational Technology, Grammar Analysis, Plagiarism Detection.},
        month = {May},
        }