Assessments of self‐organizing feature maps ( <scp>SOFM</scp> ) versus principal component analysis ( <scp>PCA</scp> ) for discriminating black gel inks

Gel pens, introduced by the Sakura Color Corporation in 1984, have become widely used in both formal and informal documentation due to their smooth writing, vibrant pigmentation, and resistance to fading. These features make them frequent subjects in forensic document examination, especially in cases involving anonymous letters, forged signatures, and disputed wills. Unlike traditional dye-based ballpoint inks, gel inks mostly are pigment-based, making them more durable and challenging to differentiate visually. This study evaluates the application of self-organizing feature maps (SOFM), an unsupervised neural network model, as an alternative to the commonly used principal component analysis (PCA), for classifying black gel pen inks using attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy data. Prior to analysis, a dissolution test was used to preliminarily group 30 black gel inks into dye-based and pigment-based categories. The FTIR spectra were pre-processed and analyzed using PCA and SOFM. While PCA achieved basic grouping based on ink type, SOFM outperformed PCA by producing 25 distinct clusters in the U-matrix, effectively discriminating both ink type and pen brand. Cross-validation of the SOFM model showed 100% classification accuracy, while testing with 20 blind samples yielded a 90% correct classification rate. These results highlight SOFM's robustness, generalization capability, and practical value in forensic casework, supporting its role as a powerful complementary or alternative chemometric tool to PCA, offering improved classification for accurate black gel ink identification.