Comparative Analysis of Traditional Machine Learning Algorithms for Text Classification in Natural Language Processing: A Study of Naive Bayes, Random Forest, and Support Vector Machine with TF-IDF Vectorization

Abstract—This paper presents a rigorous comparativeevaluation of three classical machine learningalgorithms—Naive Bayes (NB), Random Forest (RF), andSupport Vector Machine (SVM)—applied to multi-class textclassification within the Natural Language Processing(NLP) domain. TF-IDF (Term Frequency-InverseDocument Frequency) vectorization is employed for featureextraction, supplemented by Truncated Singular ValueDecomposition (SVD) for dimensionality reduction. Astructured domain-specific corpus spanning threeclasses—Artificial Intelligence (AI), Machine Learning(ML), and Natural Language Processing (NLP)—serves asthe benchmark dataset. Experimental results demonstratethat RF and SVM achieve perfect classification accuracy(100%) while NB attains 80%, exposing fundamentaltrade-offs between probabilistic simplicity anddiscriminative robustness. A novel TF-IDF featureweighting modulation framework parameterised byembedding intensity α is introduced and validated.Algorithmic complexity analyses, feature importancerankings, and a structured migration roadmap towardtransformer-augmented hybrid pipelines are provided. Thefindings yield actionable baselines for IT practitioners, NLPresearchers, and policymakers deploying scalable textclassification systems.