Fish Classification Based on Robust Features Selection Using Machine Learning Techniques

The taxonomic identification of fishes is a time-consuming process and making errors is indispensable for those who are not specialists. This system proposes an automated species identification system to identify taxonomic characters of species based on specimens. It also provides statistical clues for assisting taxonomists to identify accurate species or review misdiagnosed species. For this system, feature selection is an essential step to effectively reduce data dimensionality. By using combination theory, this system creates the set of attribute pairs to construct the training dataset. And then each attribute pair in training dataset is tested by using two classifiers. Based on the accuracy result of each classifier on attribute pairs and the majority voting of each feature in these attribute pairs, this system selects the most relevant feature set. Finally, this system applied three supervised classifiers to verify the effectiveness of selected features subset.