Multi-classifier models to improve the accuracy of fish landing application

Despite the numerous fish classification systems developed over the years, they often suffer from poor prediction accuracy, necessitating further improvement. This study addresses this issue by comparing the performance of different classifiers on fish landing datasets (2005-2019) obtained from the Department of Fisheries Malaysia (DOFM). The focus is on the East Coast of Peninsular Malaysia. The classifiers evaluated include Sequential minimal optimization (SMO), naïve Bayes (NB), multi-layer perception (MLP), instance-based for k-nearest neighbor (IBK), and random forest (RF). The performance of each classifier is assessed using classification accuracy and confusion matrix metrics, employing a 10-fold cross-validation method. Additionally, a multi-classification technique is applied to enhance the accuracy of individual classifiers and determine the most effective approach for generating an accurate dataset. The study reveals that the combinations RF+SMO+NB+MLP and SMO+RF+NB+MLP outperform single classifiers and other fusion methods, achieving the highest accuracy at 80.95%. This indicates that a multi-classifier approach can significantly enhance the performance of individual classifiers. The findings highlight the effectiveness of the multi-classifier approach in improving prediction accuracy for fish classification. The identified combinations, RF+SMO+NB+MLP and SMO+RF+NB+MLP, demonstrate superior performance and can serve as a robust methodology for fish landing classification in the context of the East Coast of Peninsular Malaysia. Further research and implementation of such multi-classifier approaches could contribute to more accurate and reliable fish classification systems.