FISH SPECIES CLASSIFICATION USING MACHINE LEARNING AND DEEP LEARNING MODELS

Abstract

Fish species classification is fundamental to ecological monitoring, biodiversity con servation, and sustainable fishery management. Conventionally, the identification of fish species is based upon manual observations by domain experts and, hence, incurs heavy expenses with time, labor, and considerable lows in scalability. This study provides broad coverage of the fish species classification problem, with a huge focus on both traditional machine learning (ML) models and architectures of deep learning (DL). A custom dataset of underwater fish images was composed, with images of nine different fish species to cater to various environmental condition-based considerations in the training and testing of several models. The ML models discussed include Support Vector Machines (SVM), Random Forests (RF), Decision Trees (DT), Logistic Regression (LR), and Naive Bayes (NB): all of these were subject to two types of dimension reduction, namely Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). These were com pared against DL models including VGG-19, DenseNet121, EfficientNet B0, Inception V3, ResNet150 V2, and LSTMs, with evaluations conducted for accuracy, precision, re call, and F1-score. The experimental results demonstrated that DL models significantly outclassed conventional ML algorithms in classification accuracy and the ability to han dle variability in images. VGG-19 attained 99.4% overall accuracy; DenseNet121 and EfficientNet B0 followed closely and are considered fit for deployment in real-world fish classification systems. Image preprocessing, normalization, and data augmentation were deemed critical in improving model performance. This study emphasizes the possibility of having deep learning automate fish species recognition with high accuracy under difficult underwater conditions. The present findings open several avenues for real-time marine surveillance, automated ecological data analysis, and smart decision support systems for marine biologists and conservationists.