Prediction of Potential Fishing Zones Using K-Means Clustering and Random Forest in Batam Waters

Abstract

Identification of potential fishing zones remains a significant challenge in fisheries management, particularly in coastal and island waters characterized by high spatial and temporal environmental variability. In Batam waters, fishing activities are still dominated by fishermen's experience and heuristic judgment, while existing studies often focus on a single prediction model or limited environmental parameters. This indicates a research gap, namely the lack of an integrated framework that simultaneously captures environmental heterogeneity and improves prediction accuracy using a data-driven approach. To address this gap, this study proposes a hybrid data mining framework that explicitly integrates unsupervised environmental zoning and supervised classification for predicting fishing potential. Weather and oceanographic variables—including sea surface temperature, chlorophyll-a concentration, wind speed, ocean current speed, and salinity—are used in conjunction with historical fish catch data. K-Means clustering is first used to identify homogeneous marine environmental zones, which are then incorporated as contextual features into a Random Forest classification model. Model performance is then evaluated using accuracy, precision, recall, F1 score, and confusion matrix analysis. The results show that the proposed hybrid approach achieves an accuracy of 89.2% and an F1 score of 89.1%, representing a quantitative improvement of approximately 5.6% in accuracy and 5.0% in F1 score compared to the baseline Random Forest model without clustering. This comparison clearly demonstrates that the integration of clustering information significantly improves classification performance. Furthermore, feature importance analysis confirms that sea surface temperature and chlorophyll-a concentration are the most influential predictors, while cluster labels contribute indirectly by improving the model's contextual understanding of complex environmental conditions. The novelty of this research is articulated through the integration of unsupervised marine environmental zoning with supervised machine learning in a local fisheries context, which allows for improved predictive performance and enhanced model interpretability. Unlike conventional approaches that treat environmental variables independently, the proposed framework captures multidimensional environmental interactions in a structured manner. The implications of these findings are profound. The proposed model can support data-driven decision-making for fishermen by reducing search time and operational costs, while providing a scientific basis for fisheries managers for spatial planning and sustainable resource management. Therefore, this research contributes both methodologically and practically to the advancement of intelligent fisheries prediction systems in dynamic coastal environments such as Batam waters.