Principal Component-based Modeling Approaches for Predicting Soil Organic Matter

Authors

  • Kamolchanok Panishkan Department of Statistics, Faculty of Science, Silpakorn University, Nakhon Pathom 73000, Thailand.
  • Natdhera Sanmanee Department of Environmental Science, Faculty of Science, Silpakorn University, Nakhon Pathom 73000, Thailand.
  • Sirikanlaya Pramual Mathematics Program, Faculty of Science and Technology, Sisaket Rajabhat University, Sisaket 33000, Thailand.

Keywords:

feed forward artificial neural networks, genetic algorithm, multiple linear regression, principal components, soil organic matter

Abstract

The objective of this research is to study three statistical modeling approaches; namely stepwise multiple linear regression, a feed-forward artificial neural network and a genetic algorithm for predicting quantity of organic matter in soil. Soil samples were selected from three fruit farming agricultural areas in the western region of Thailand; Nakhon Pathom, Samut Sakhon and Samut Songkram. Seventeen soil properties were measured on the soil samples and are used as original variables. To reduce the number of original variables and eliminate data collinearity, a principal component analysis was applied. The models were based on the first five principal components which accounted for 75.81% of total variance. Model performance was measured by performance indexes which are IA, RMSE, MBE and MAE. The results of this study indicated that the genetic algorithm model performs the best among these three models in a validation step and is the most efficient model to predict soil organic matter.

Downloads

How to Cite

Panishkan, K., Sanmanee, N., & Pramual, S. (2015). Principal Component-based Modeling Approaches for Predicting Soil Organic Matter. Thailand Statistician, 9(1), 51–64. Retrieved from https://ph02.tci-thaijo.org/index.php/thaistat/article/view/34287

Issue

Section

Articles