Correlation of environmental variables with Heteromastus filiformis habitat density: Gap-filling and machine learning approaches

Benthic invertebrates have been used as ecological indicators to assess marine ecosystem health. However, the intricate interactions between these organisms and their environment complicate detailed analyses of the effects of environmental disturbance events. This study investigates the impact of various environmental variables on the habitat density distribution of Heteromastus filiformis using gap-filling and machine learning approaches. Data collected from the Korea Strait, Yellow Sea, and East Sea were analyzed to forecast habitat density using various machine learning models combined with different gap-filling methods and feature selection. The results show that the artificial neural network model with a multilayer perceptron architecture outperforms others, achieving a weighted mean absolute error of 0.865 and a weighted accuracy of 0.631 in predicting Heteromastus filiformis habitat density when combined with station-wise correlated feature gap-filling method and the selection of 16 environmental features. The model also forecasts that simultaneous increases in ocean salinity and water depth will have the most detrimental impact on its habitat density compared to other environmental features. This study demonstrates the feasibility and effectiveness of machine learning techniques in bridging data gaps and enhancing the understanding of influential factors in benthic marine ecosystems. • Gap-filling methods reliably reconstructed missing data from 155 coastal stations. • ANN with SCF imputation and feature selection gave the highest prediction accuracy. • Salinity and water depth were the strongest negative drivers of habitat density. • Sediment sorting and silt had positive effects, while excess OM reduced densities. • ML–ecology links provide insights for monitoring and management under human impacts.