Digital color analysis and machine learning for ballpoint pen ink clustering and aging investigation

Fraudulent activities often involve document manipulation, which poses a significant challenge to forensic science. To address this issue, a novel method was developed that combines intended artificial UV pre-degradation, digital color analysis (DCA) of stroke images, and various machine learning (ML) models. This method can cluster blue ballpoint pen inks and predict their photodegradation time. The results of the study indicate that the k-shape clustering method is highly effective in differentiating between inks based on their degradation curve patterns and HSV or RBS color features, aligning well with results from chromatography analyses. Furthermore, the random forest regression model demonstrated superior performance in predicting age, exhibiting the highest coefficients of determination. The DCA-ML method is a straightforward, cost-effective, and highly accurate solution for clustering blue pen inks. Using photodegradation curves to predict document age could eliminate the need for conventional physicochemical analysis techniques. [Display omitted] •Identified distinct aging patterns for different blue ballpoint pen ink samples.•Developed the workflow for processing the obtained images based on ML.•Found two best feature sets (HSV, RBS) for clustering pen inks.•Proposed clustering aligned well with the chemical-based approach involving DCA.•Shown the superiority of the random forest regression model (R2 0.993–0.996).