Machine learning for forewarning crop diseases.
Rajni Jain, Sonajharia Minz, Ramasubramanian V.
Author Affiliation: National Centre for Agricultural Economics and Policy Research, New Delhi, India.
Journal of the Indian Society of Agricultural Statistics 63 : 97-107
Abstract : With the advent of computers, the development of accurate forewarning systems for incidence of crop diseases has been increasingly emphasized. Timely forewarning of crop diseases will not only reduce yield losses but also alert the stakeholders to take effective preventive measures. Traditionally, Logistic Regression (LR) and discriminant analysis methods have been used in forewarning systems. Recently, several machine learning techniques such as decision tree (DT) induction, Rough Sets (RS), soft computing techniques, neural networks, genetic algorithms etc. are gaining popularity for predictive modelling. This paper presents the potential of three machine learning techniques viz. DT induction using C4.5, RS and hybridized rough set based decision tree induction (RDT) in comparison to standard LR method. RS offers mathematical tools to discover hidden patterns in data and therefore its application in forewarning models needs to be investigated. A DT is a classification scheme which generates a tree and a set of rules representing the model of different classes from a given dataset. A java implementation of C4.5 (CJP) is used for DT induction. A variant of RDT called RJP, combines merits of both RS and DT induction algorithms. Powdery mildew of Mango (PWM) is a devastating disease and has assumed a serious threat to mango production in India resulting in yield losses of 22.3% to 90.4%. As a case study, prediction models for forewarning PWM disease using variables viz. temperature and humidity have been developed. The results obtained from machine learning techniques viz. RS, CJP and RJP are compared with the prediction model developed using LR technique. The techniques RJP and CJP have shown better performance over LR approach.