Environmental Studies Journal
Volume 4, No. 1, 2025
Pages 86-115
DOI: 10.36108/esj/5202.40.0150
Predictive Machine Learning Modelling of Biological Indices from Physico-Chemical Characteristics in Sand Dredging Areas of Lagos Lagoon
*AJAYI Aderemi Pius, SHOGA Aramide Mutiat, OSHONEYE Jacob oluwaseyi, ADEYEMO Lateef Taiwo
1Environmental toxicology and pollution management unit, Zoology Department, University of Lagos Akoka, Yaba, Lagos, Nigeria
*Corresponding Author- Email: remipius@yahoo.com Orcid No: https://orcid.org/0009-0006-3090-9292
Abstract
Sand dredging in the Lagos Lagoon Complex poses significant threats to aquatic ecosystem integrity. This study applied three machine learning (ML) regression algorithms; Random Forest (RF), K-Nearest Neighbour (KNN), and Adaptive Boosting (AdaBoost) to model and predict biological indices from physicochemical water quality variables across four sampling locations (Ibese, Majidun, Makoko, and Ilaje) in active and reference sand dredging zones of the Lagos Lagoon. Physicochemical parameters, antioxidant enzyme activities (CAT, SOD, GST, GPx) and MDA, a lipid peroxidation marker in Chrysichthys nigrodigitatus and Ethmalosa fimbriata, and biological diversity indices (Shannon–Wiener H’, BMWP) derived from phytoplankton and benthic macrofauna were assessed. The VIF screening reduced 13 physicochemical predictors to four non-collinear variables: dissolved oxygen (DO), total dissolved solids (TDS), nitrate, and phosphate. The RF achieved the strongest prediction of benthic diversity (H’ Benthos: R² = 0.544, RMSE = 0.278), while AdaBoost outperformed RF and KNN for H’ Combined (R² = 0.412, RMSE =0.336) and BMWP (R² = 0.439, RMSE =2.648). TDS was the dominant predictor across all biological targets (45–84%), identifying ionic loading and oxygen dynamics as the principal drivers of biological community structure. Canonical Correspondence Analysis confirmed pH and conductivity as the primary environmental axes separating locations, with Makoko exhibiting severely elevated MDA in both fish species. Piecewise regression identified DO thresholds of 4.47–6.41 mg/L and TDS breakpoints of approximately 290–318 mg/L as critical ecological limits with 95% CI. These thresholds are proposed as regulatory benchmarks for sustainable dredging-associated environmental stress management in the Lagos Lagoon.
Keywords: Antioxidant enzyme, Biological diversity indices, Lagos Lagoon, Physicochemical parameters, Sand dredging
