IMPROVING MACHINE LEARNING MODEL PERFORMANCE FOR NILE TILAPIA FCR PREDICTION VIA FUNCTIONAL ADDITIVE CATEGORIZATION: A STEP-WISE FEATURE ADDITION ANALYSIS

Abstract

The purpose of this research is to compare the performance of machine learning models and identify the optimal input dataset using step-wise feature addition analysis to predict feed conversion ratio (FCR) in Nile tilapia. We collected a total of 171 diet formulations from studying the effect of dietary supplementation with functional additives on the growth performance of Nile tilapia. The functional additives were classified in 3 groups, i.e., probiotics, prebiotics, and plant extracts. In this research, we studied 8 machine learning models, including Multiple Linear Regression, Ridge, Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost. This study found that the CatBoost achieved the best accurate FCR predicted performance (RMSE = 0.0889 and R² = 0.9011). Additionally, the step-wise feature addition analysis found that identifying types and dosages of functional additives significantly improved FCR prediction performance. This research shows the enhancing performance of machine learning to support precision aquaculture.

Keywords: Machine Learning, Step-wise Feature Addition Analysis, Feed Conversion Ratio, Functional Additives