ESTIMATION OF FISH CATCHES USING MACHINE LEARNING METHOD (CASE STUDY: PPN TANJUNG PANDAN, BANGKA BELITUNG ISLANDS) Destika Agustina Widiawan & Susanna
Institut Teknologi Bandung
Abstract
Fish catches are very important data for monitoring Indonesian marine products. Technological developments related to current data can be used to predict fish catches using Machine Learning methods. Fishing results at PPN Tanjung Pandan Port are influenced by social factors and oceanographic factors. Data on excessive fish and several social factors such as the use of oil, kerosene, diesel, and fishing logistics from the PPN Tanjung Pandan Port Annual Report are combined with remote sensing data for oceanographic factors such as SST, chlorophyll-a, salinity, wind and currents from 2008-2017 used as data to predict fishing with the Random Forest and XGBoost methods. The correlation between oceanographic parameters of salinity and fish catches is the highest at 0.35 compared to other oceanographic parameters. The social correlation faktor for the highest fish catches was ship visits at 0.34- oil(liter) at 0.29- and sugar at 0.22. The prediction results for fighting fish in PPN Tanjung Pandan for the XGBoost method were better than the Random Forest prediction for all the best fish models, XGBoost with 3-year data with an accuracy of 0.86- the best model for prediction of Trevallies fish is XGBoost with 5 years of data with an accuracy of 0.55- and the prediction of the best Short bodied mackerel model is XGBoost with 10 years of data with an accuracy of 0.40.
Keywords: Fish catches, PPN Tanjung Pandan, Machine Learning
