アグロテクノロジー

概要

Performance Analysis of Regression-Machine Learning Algorithms for Predication of Runoff Time

Marwan Khan and Sanam Noor

Worldwide 70 percent of water is used in agriculture practices, in which 50% of water is lost due to improperly planned and inefficient irrigation system. Precision irrigation system has long been used on individual farms scale. Very rare work has been done so far to utilize the excessive irrigation water of one farm in another farm. In this research, we address the problem of predicting the runoff time between two farms. We propose Runoff time model which accepts irrigation depth, soil moisture and crop stage (CN) and time of concentration as input parameters and estimate runoff time. Machine learning algorithms i.e., Multiple Linear Regression (MLR), Artificial Neural Network-Levenberg Marquardt (LMA-ANN), Decision Trees/Regression Tree (DT/RT) and Least Square Support Vector Regression (LS-SVR) have been used for learning and predication purposes. A comparison has been made among these algorithms to choose best algorithm for irrigation runoff time prediction. Experimental results show that regression tree aces the results in terms of highest R-square value, lowest Mean square error. While MLR shows the worse result in terms of least R-square value, highest means square error. The algorithms Regression tree is ranked first-outstanding, ANN-LMA is ranked second-good, LS-SVR is ranked third-fair and MLR is ranked last-poor on the basis of the regression error metrics/ performance evaluation parameters. Hence it is strongly suggested that regression tree is an ideal machine learning-regression algorithm to be deployed on the Wireless Sensor Network (WSN) node for the predication of runoff time.