Classification models based on the gonadosomatic index to determine gonadal maturity stages: a case study in the Peruvian anchovy Engraulis ringens

Abstract

Accurate classification of gonadal maturity stages is essential to ensure reliable estimates of reproductive indicators. In Peruvian anchovy (Engraulis ringens), two gonadal maturity scales have been used: the empirical scale proposed by Einarsson et al. (1966) and the histologically validated scale developed by Buitrón et al. (2015). The adoption of this scale underscores the need to harmonize or retrospectively adjust historical maturity records, which is essential for analysing long-term changes in anchovy reproductive dynamics. In this study, we evaluate the efficacy of the gonadosomatic index (GSI) in predicting stages and groups of gonadal maturity. Data from 115 research surveys conducted between 1984 and 2022 were analysed. Classification models using different algorithms were applied, with the random forest algorithm demonstrating the highest accuracy and robustness, achieving a mean accuracy of >90% and a logarithmic loss of 0.13. GSI and total length were identified as the most influential predictors of gonadal maturity. The classification of historical data (1984–2014) using the M-5 model (which includes GSI, total length, latitude, distance from the coast and condition factor) demonstrated high concordance with the reproductive patterns observed in recent periods (2015-2022), supporting its predictive reliability. The results indicate that the combined use of GSI and machine learning-based classification models is a valuable tool for identifying gonadal maturity stages from historical data and harmonizing them with the current maturity scale, facilitating the interpretation of variations in the reproductive behaviour of the resource over time.