COMPARISON OF LINEAR REGRESSION AND ARTIFICIAL NEURAL NETWORK MODELS FOR PREDICTING FISH CATCH VOLUME IN URENG VILLAGE, CENTRAL MALUKU

Abstrak

This study aims to develop a predictive model for fish catch volume in Ureng Village, Central Maluku, using a mathematical modeling approach based on artificial intelligence with the Scikit-Learn and TensorFlow libraries. The research dataset consists of 24 monthly data records collected from July 2024 to June 2025. The data were obtained through a combination of primary and secondary collection methods. Primary data were gathered through interviews, field observations, and fishermen’s catch records, while secondary data included oceanographic parameters such as sea surface temperature, weather conditions, and current velocity. Two main models were developed: a linear regression model using Scikit-Learn as the baseline and a neural network model using TensorFlow as the comparator, both trained and evaluated on the same dataset to ensure consistency. The testing results show that the linear regression model produced a Mean Squared Error (MSE) of 0.8821 and a coefficient of determination (R²) of 0.682, while the neural network model achieved an MSE of 0.5423 and an R² of 0.815. These findings indicate that the neural network model is more capable of capturing nonlinear patterns among temperature, weather, and current variables, resulting in higher prediction accuracy than the linear model. Nevertheless, this study is limited by the relatively small sample size and the need for a more detailed description of the data period and measurement units to allow a more objective evaluation of the model’s performance. Overall, this AI-based approach has the potential to support more efficient, adaptive, and sustainable decision-making in fishery planning for coastal communities.