SCALABLE FISH LARVAE GROWTH PREDICTION USING DEEP LEARNING AND PCA-BASED FEATURE EXTRACTION

This study focuses on predicting fish larvae growth using advanced machine learning techniques. According to recent aquatic biology reports, over 70% of larval growth variability remains unexplained by traditional models, impacting ecological assessments and fisheries management. Additionally, datasets often contain missing or inconsistent values, with up to 15% of records having incomplete features, complicating analysis. Conventional manual approaches for growth prediction rely heavily on domain expertise and simple statistical models, which struggle to capture complex nonlinear relationships inherent in biological data. These manual methods are time-consuming, prone to human bias, and often fail to generalize well across diverse environmental conditions. To address these challenges, the proposed method integrates a hybrid modeling framework combining KNearest Neighbors (KNN), Support Vector Regression (SVR), and a Convolutional Neural Network (CNN) with Principal Component. Analysis (PCA)-based dimensionality reduction. This novel approach preprocesses raw data with label encoding and feature scaling, extracts essential components through PCA to reduce dimensionality, and feeds these into tailored regression models. The CNN model specifically leverages 1D convolutional layers to capture temporal and spatial feature dependencies, while KNN and SVR provide complementary predictions based on neighborhood and margin-based learning. This ensemble harnesses the strengths of each method, leading to improved prediction accuracy and robustness, thereby overcoming limitations of existing manual techniques and offering a scalable, automated solution for ecological growth modeling