ABSTRACT: |
The environmental factors,
soil and productivity mapping of a certain culture produces a great amount
of data, which may contain important information to be used in the decision
making processes concerning actions in the field. The extraction of implicit
and potentially useful information from databases is the main objective
of the area denominated Data Mining. This area uses Artificial Intelligence
techniques such as Decision Tree and Genetic Algorithms, in order to accomplish
its tasks. The objective of this work was to compare results obtained
through those techniques in data mining about soil physiochemical characteristics
and soy productivity, obtained experimentally. The Decision Tree and Genetic
Algorithm were implemented in Prolog and Borland Delphi® Professional
languages, respectively. The results were presented in the form of production
rules, being the goal to obtain rules to predict productivity indexes
above 2t/ha, 100% reliable. The rules generated by the algorithm of the
Decision Tree used the operators < or >= to relate a certain value
to each of the analyzed attributes. The Genetic Algorithm, due to the
easiness in handling continuous values, enabled the use of more operators,
also accepting the adoption of values interval for an attribute in the
rules. As for the composition of the rules, AD presented larger variety
of attributes, while in AG the variability was noticed more specifically
in the values, once the rules pondered around certain attributes, considered
the most important. Therefore, the conclusion is that the Genetic Algorithm
enabled better treatment of the data due to the diversity of operators,
global search and pre-process simplification by manipulating continuous
values. Keywords: Genetic Algorithm, Decision Tree, soil physical-chemical properties |