Artificial Neural Network Performance on Pakcoy Leaf Fresh Weight Model

Abstract

The growth of leaf biomass can be predicted from an increase in the surface area and thickness of the leaves. Measurements of leaf biomass are approached with the fresh weight of the leaves. The relationship between biomass and leaf surface area commonly performed by regression analysis. The analysis requires assuming linear relationship between dependent variables and independent variables. Artificial Neural Network (ANN) is alternative that can be used to analyze the relationship of leaves and leaf biomass without requiring linear relationships. The research aimed to evaluate ANN performance in determining the fresh weight of pakcoy leaves based on leaf area parameters. Datasets in the study included leaf area datasets and length-width datasets. ANN architecture used Multi Layer Perceptron (MLP) with backpropagation. Ramsey’s test results showed that leaf area datasets is linier model and length-width datasets is nonlinier model. ANN performs well in predicting leaf fresh weight data on both nonlinear and linear models. The best ANN architecture for modeling the leaf fresh weight with leaf area is MLP (1-3-1) while the leaf fresh weight model with length and width is MLP (2-3-1).