Application of Logistic Regression and Numerical Methods for Automated Bot Detection in Social Media Networks

Abstract

The frenzied exponential growth of social media sites has made them rich hunting ground for bot infestation, presenting formidable challenges to information purity, public discourse, and web security. In this study, the joint application of Logistic Regression, a conventional supervised learning algorithm, with numerical methods is used to design a statistically robust and computationally efficient scheme for automated bot detection. Conventional methods based on heuristics tend not to include subtle patterns because they rely on fixed rule sets. Our method, however, adjusts to feature fluctuations between platforms like Facebook and Twitter by applying a continuous variable model and iterative optimization processes. Using real-world datasets (e.g., Botometer, Twint-processed Twitter data, and Kaggle social bots dataset), we implement stepwise logistic modeling using convergence-enhancing numerical algorithms such as Newton-Raphson iteration and stochastic gradient descent. Numerical validation verifies the predictive capability of the model with an F1-score of over 93%, significantly higher compared to common classifiers. The study indicates that the integration of statistical modeling and deterministic numerical methods can render detection improved, false positives reduced, and interpretability enhanced in high-dimensional data environments.