Aquilaria essential oil species discrimination using K-NN model applied to GC-MS coupled with GC-FID data

The classification of Aquilaria essential oil, widely referred to as agarwood oil, presents a challenge due to its chemically diverse composition and the subjectivity associated with traditional species identification methods. This study introduces a machine learning approach for species discrimination by analyzing chemical compounds obtained through gas chromatography-mass spectrometry (GC-MS) coupled with gas chromatography-flame ionization detection (GC-FID). The research focuses on four Aquilaria species, namely Aquilaria beccariana, Aquilaria malaccensis, Aquilaria crassna and Aquilaria subintegra, with the aim of developing a reliable classification framework using the k-Nearest Neighbour (k-NN) algorithm. A total of thirty-eight chemical compounds was identified across all samples, with six consistently detected in all species. These six were subsequently narrowed to four significant compounds using a structured pre-processing workflow comprising cross-correlation analysis to assess compound consistency, boxplot analysis to visualize interspecies variation and selection frequency to prioritise compounds appearing repeatedly across species. Based on statistical strength and visual consistency, four significant compounds, namely dihydro-β-agarofuran, δ-guaiene, 10-epi-γ-eudesmol and γ-eudesmol, were selected for model development due to their high discriminatory power. These compounds were tested in four pairings using the k-NN model under four distance metrics (Euclidean, Minkowski, Correlation and Spearman). The Euclidean and Minkowski distance measures achieved 100% discrimination accuracy, with the pairings of δ-guaiene with 10-epi-γ-eudesmol and 10-epi-γ-eudesmol with γ-eudesmol performing most consistently. The findings showed that Aquilaria subintegra and Aquilaria beccariana were most clearly distinguished, while Aquilaria malaccensis and Aquilaria crassna exhibited partial similarity but were still reliably identified through optimal pairings. The model was rigorously evaluated using classification metrics including accuracy, sensitivity, specificity, precision and confusion matrices, with Euclidean and Minkowski achieving 100% accuracy while Correlation and Spearman recorded below 63%. This research presents a validated framework for Aquilaria oil species discrimination based on chemical profiling and intelligent modelling, offering practical potential for automated authentication systems and demonstrating applications of electrical engineering in signal analysis, pattern recognition and smart classification technology.