Multi-Dimensional Benchmarking of Text Classification Models
Year: 2026
Type: conference-paper
Abstract: A thorough benchmarking system for text classification models is presented in this work, which assesses performance in a number of areas, such as scalability, accuracy, robustness, and computing economy. We preprocess text using TF-IDF vectorization with a controlled vocabulary using the popular 20 Newsgroups dataset, and then evaluate a variety of models, from classical approaches like Logistic Regression, Decision Trees, and Support Vector Machines to ensemble approaches like Random Forest, XGBoost, and deep neural networks. Our assessment includes scalability studies, robustness testing using Gaussian noise, resource profiling (estimates of CPU, memory, and energy), and hyperparameter sensitivity analysis. The findings show that while more sophisticated approaches, such as ensemble and deep learning models, achieve high performance at the expense of higher computational costs and decreased robustness under noise, more conventional models, such as SVM and logistic regression, offer competitive accuracy with lower resource demands. These results draw attention to important trade-offs and highlight the need to customize model selection for certain deployment scenarios. In order to further refine model selection for realistic, largescale applications, future work will concentrate on automated hyperparameter optimization, expansion to different datasets, real-time energy monitoring, and advanced robustness testing.
Authors: Shaikh Amra Bano, Surya Prakash Mishra
Institution: Sam Higginbottom Institute of Agriculture
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201834145