Lack of data: Is it enough estimating the coffee rust with meteorological time series?

Abstract

Rust is the most economically important coffee disease in the world. Coffee rust epidemics have affected a number of countries including: Colombia, Brazil and Central America. Researchers try to predict the Incidence Rate of Rust (IRR) through supervised learning models, nevertheless the available IRR measurements are few, then the data set does not represent a sample trustworthy of the population. In this paper we use Cubic Spline Interpolation algorithm to increase the measurements of Incidence Rate of Rust and subsequently we construct different subsets of meteorological time series: (i) Daily meteorology, (ii) Meteorological variation, and (iii) Previous meteorology using M5 Regression Tree, Support Vector Regression and Multi-Layer Perceptron. Previous meteorology with Multi-Layer Perceptron have shown better results in measures as Pearson Coefficient Correlation of 0.81 and Mean Absolute Error = 7.41%.