Implementing Machine Learning Methods for Ballpoint Pen Ink Classification based on Mass Spectrometry Data: Toward a Forensic Application

Abstract

Mass spectrometry (MS) is widely used for material analysis in various applications including forensic science. This work explores computational techniques and develops an application called "MSpec" using suitable algorithms for extracting informative parts of the MS dataset that aims towards pen ink classification. The system is intended as a tool that is capable of giving preliminary answers for such forensic analyses of documentary evidence involving different pen-ink types on writing. Support Vector Machine (SVM) was implemented and compared with other machine learning techniques via systematic performance assessments. They were trained and tested using MS data acquired from 10 blue-ink ballpoint pen samples, which were pre-processed using optimized steps. The results show that the tested models performed well in classifying the pen ink samples, with the SVM cubic kernel model giving the highest accuracy of 96.0%. Furthermore, dimensionality reduction of the dataset through peak detection helps improve the classification accuracy.