DATA MINING FOR THE ASSESSMENT OF MANAGEMENT AREAS IN PRECISION AGRICULTURE

Abstract

ABSTRACT Precision Agriculture (PA) uses technologies with the aim of increasing productivity and reducing the environmental impact by means of site-specific application of agricultural inputs. In order to make it economically feasible, it is essential to improve the current methodologies as well as proposing new ones, in which data regarding productivity, soil, and compound indicators are used to determine Management Areas (MAs). These units are heterogeneous areas within the same region. With these methodologies, data mining (DM) techniques and algorithms may be used. In order to integrate DM techniques to PA, the aim of this study was to associate MAs created for soy productivity using the Fuzzy C-means algorithm by SDUM software over a 9.9-ha plot as the reference method. It was in opposition to the grouping of 2, 3, and 4 clusters obtained by the K-means classification algorithms, with and without the Principal Component Analysis (PCA), and the EM algorithm using chemical and physical data of the soil samples collected in the same area during the same period. The EM algorithm with PCA modeling had a superior performance than K-means based on hit rates. It is noteworthy that the greater the number of analyzed MAs, the lower the percentage of hits, in agreement with the result shown by SDUM, which shows that two MAs compose the best configuration for this studied area.

Keywords

• algorithms
• EM
• KDD
• K-means
• Weka