Optimizing forensic file classification: enhancing SFCS with β_k hyperparameter tuning.

Area of Science:

• Computer Science
• Information Science
• Forensic Science

Background:

• Traditional topic modeling parameters (α, β, βj) in forensic analysis often lead to suboptimal topic distribution, sparsity, and overfitting.
• Existing methods struggle with data skewness and noise from polysemic word pairs, limiting classification model convergence.
• Current topic modeling approaches can result in inefficient classification with high time complexity.

Purpose of the Study:

• To propose the SDOT Forensic Classification System (SFCS) to address limitations in forensic topical modeling.
• To introduce a new functional parameter (βk) for identifying seed words based on semantic and contextual similarity.
• To enhance the accuracy and efficiency of file classification in forensic contexts.

Main Methods:

• Developed the SDOT Forensic Classification System (SFCS) incorporating a novel parameter βk.
• Employed semantic and contextual similarity of word vectors to identify seed words.
• Integrated hyperparameter optimization and hyperplane maximization for model refinement.

Main Results:

• The SFCS successfully removed 278,000 irrelevant files and identified 5,600 suspicious files using 700 blacklisted keywords.
• Achieved a file classification accuracy of 94.6%, with 94.4% precision and 96.8% recall.
• Reduced time complexity to O(n log n) through optimized parameter integration.

Conclusions:

• The SFCS, with the functional parameter βk, effectively compels topic distribution to model curated seed words, generating pertinent topics.
• The proposed system significantly improves the identification of relevant and suspicious files in forensic corpora.
• SFCS demonstrates superior performance in accuracy, precision, and recall with enhanced computational efficiency for forensic file classification.