Abstract: In order to understand the morphological characteristics of sagittal otoliths of fish species in Carangidae and to study the effectiveness of different machine learning algorithms on the population classification of Carangidae in the South China Sea, otolith samples of Selaroides leptolepis and Decapterus macrosoma were collected from the South China Sea. The otoliths were measured with 4 basic morphological parameters and transformed into 6 morphological indexes, and the differences in otolith morphology were compared between two species, forty four elliptic Fourier descriptor coefficients were extracted from the otoliths for principal component analysis (PCA), and four different machine learning classification models, namely, linear discriminant analysis (LDA), random forest(RF), K-nearest neighbor(KNN), and support vector machines(SVM) were used to discriminate them. The result showed that the otolith parameters and body length of the two species were significantly different. The length, height, area, perimeter, and body length of the two species were all power functions. From the analysis of morphological indexes, the otoliths of S. leptolepis had a lower ring rate described as more round and more regular, there were significant differences in morphological indexes between the two groups. PCA showed that the first and second principal component accounted for 20.1% and 13.3% of the total variation, respectively. Among the four classification models, the highest correct rate of RF was 100%, and the lowest was 93.3% for SVM.

Key words: otolith morphology, Selaroides leptolepis, Decapterus macrosoma, machine learning, discriminant analysis, South China Sea