Neural Network in Modeling Malaysian Oil Palm Yield

Ismail Ismail

doi:10.3844/ajassp.2011.796.803

Abstract

Problem statement: Forecasting of palm oil yield has become an important element in the management of oil palm industry for proper planning and decision making. The importance of yield forecasting has led us to explore modeling of palm oil yield for Malaysia using the most recent development of Artificial Neural Network (ANN). The main issue in yield forecasting is to predict the future value with the minimum error. Approach: Artificial neural networks are computing systems containing many interconnected nonlinear neurons, capable of extracting linear and nonlinear regularity in a given data set. It is an artificial intelligence model originally designed to replicate the human brain’s learning process, a network with many elements or neurons that are connected by communications channels or connectors. The ANN can perform a particular function when certain values are assigned to the connections or weights between elements. In this study, a secondary data set from the Malaysian Palm Oil Board (MPOB) on the foliar nutrient composition, fertilizer trials and Fresh Fruit Bunch (FFB) yield were taken and analyzed. The foliar nutrient composition variables are the nitrogen N, phosphorus P, potassium K, calcium Ca and magnesium Mg concentration, while the fertilizer trials data are the N, P, K and Mg fertilizers and are measured in kg per palm per year. The foliar composition data was presented in the form of measured values whiles the fertilizer data in ordinal levels, from zero to three. Results: Two experiments were conducted to demonstrate the implementation ANN and for both experiment, the result demonstrated that the number of hidden nodes produces an effect to the overall forecast performance of the ANN architecture. From the first experiment, it shows that the number of runs does not affect the ANN performance, but changing the momentum to learning rates, due to shows a significant improvement in the forecast result. The experimental result will be in the form of statistical analysis, the best neural network performance, the residual analysis and the effect on the learning rate on the NN performance. Conclusion: This study showed that modeling of oil palm yield using neural network requires data to be prepared or modified to satisfy the requirement of the parameters involved. This analysis yields the conclusion that only the number of hidden nodes has a significant influence on the NN performance and there is no effect resulting from the number of runs or the momentum term value on the neural network’s performance.

Highlights

Neural network or popularly known as Artificial Neural Networks (ANN), are computational models that consist of a number of simple processing units which communicate by sending signals to each other over a large number of weighted connections
The performance of NN was due to the effect of the combination activation function in the hidden layer and output layer
The F value for number of runs, 1.6950 (p = 0.1330 and df = (5, 1914)) and the momentum term, 1.3300 (p = 0.2630 and df = (3, 1916)), show that both factors did not influence the overall performance of the neural networks. This analysis yields the conclusion that only the number of hidden nodes has a significant influence on the NN performance and there is no effect resulting from the number of runs or the momentum term value on the neural network’s performance

Summary

Introduction

Neural network or popularly known as Artificial Neural Networks (ANN), are computational models that consist of a number of simple processing units which communicate by sending signals to each other over a large number of weighted connections. One very important feature in ANN is in its adaptive nature where “learning by example” replaces “programming” in solving problems. This feature renders computational models very appealing in applications where one has little, or an incomplete understanding, of the problems to be solved, but where training data is available. Some of the most traditional applications include the area of Classification- to determine military operations from satellite photographs; to distinguish among different types of radar returns (weather, birds, or aircraft); to identify diseases of the heart from electrocardiograms; Noise reduction-to recognize a number of patterns (voice, images) corrupted by noise and Prediction -to predict the value of a variable, given historic values. Examples include forecasting of various types of loads, market and stock forecasting and weather forecasting (Kubde and Bansod, 2010; Adeli and Panakkat, 2009; Wang et al, 2009; Faraway and Chatfield, 1998)

Methods

Results

Conclusion