Dr Branislav Radomirovic
2. мар 2026.
How a unique optimization of Logistic Regression is solving the "false positive" problem and outperforming state-of-the-art email filters.
Email spam is more than just a nuisance; it’s a constant interruption that costs time and compromises security. While machine learning is currently the gold standard for catching these unwanted messages, current systems still struggle with two major issues: false positives (valid mail marked as spam) and leaked spam (junk that looks legitimate).
The Innovation: Social Network Search (SNS)
Traditional machine learning methods are effective, but they often lack the fine-tuning required for modern, high-dimensional data. This research proposes a unique solution: training a Logistic Regression model using improved Social Network Search (SNS) metaheuristics.
By mimicking the way information and influence spread through social networks, this optimization technique allows the model to "search" for the best possible parameters more efficiently than standard algorithms.
High-Performance Results
To prove the model's worth, it was tested against the CSDMC2010 dataset, a well-known benchmark for high-dimensional spam data. The results were clear:
Efficiency: The model handles complex, high-degree data without slowing down.
Accuracy: It achieved higher classification accuracy when compared to existing state-of-the-art detection methods.
Reliability: The system significantly reduced the frequency of mislabeled emails.
Why This Matters for Users
By refining the "brain" of the spam filter, this approach ensures that your inbox remains a productive space. It addresses the inadequacies of older methods by being more adaptive to the sophisticated tactics used by modern spammers.
For developers and researchers, this study illuminates the power of combining metaheuristic search patterns with classic statistical models like Logistic Regression to achieve superior results in cybersecurity.