A Support Vector Machine-Based Model for Detecting Malicious Social Media Posts
The proliferation of social media platforms has led to a significant increase in the spread of malicious content, including fake news, hate speech, malware-laden links, and other offensive material capable of inflicting real harm on users and on society at large. This study proposes a Support Vector Machine (SVM) based model for the binary detection of malicious posts, in which a post is treated as malicious if it carries fake (deceptive) news content and as non-malicious if it carries real news. The publicly available Kaggle Fake News detection dataset, comprising 4,009 news articles (2,137 fake and 1,872 real) described by their URL, headline and body text, was used. Each text attribute was converted to a numeric code, and the data were partitioned into 3,207 training and 802 test samples using a random 80:20 split (random seed 10). A Scikit-learn SVC classifier with a radial basis function (RBF) kernel and default hyperparameters was trained in Python 3.7.4. On the held-out test set, the model achieved an accuracy of 88.9% (error rate 11.1%), with a macro-averaged precision of 0.889, recall of 0.890 and F1-score of 0.889; for the malicious class, precision was 0.914 and recall 0.873. These results describe performance on this dataset and split only. The study discusses the limitations of the adopted feature representation and recommends larger, multi-platform datasets in future work.
Authors
- Ayodeji Olusegun Akinwumi (ORCID: https://orcid.org/0000-0003-3080-2779)
- Olatunde David Akinrolabu (ORCID: https://orcid.org/0009-0004-3337-0240)
- Moyinoluwa Adedoyin
- Avoseh Emmanuel Deyon
- Olanrewaju Adeniyi (ORCID: https://orcid.org/0009-0006-6746-2148)
Institutions
- Adekunle Ajasin University (NG)
Publication Details
- Journal
- Interdisciplinary Journal of AI Machine Learning & Data Science
- Published
- 2026-11-22
- DOI
- https://doi.org/10.66261/ffw9gf40
- Primary Topic
- Misinformation and Its Impacts
- Type
- article
- Field-Weighted Citation Impact
- 0.00