Knowledge Based Reasoning for Agricultural Crop Management Decisions: An Experiment Using Rule Based and Artificial Neural Network Approach
No Thumbnail Available
Date
2009-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Studies conducted so far and annual reports frequently issued by the national bank of
Ethiopia indicated that agriculture remained the main source of living for the greatest portion
of Ethiopia's population. Despite the fact that government is giving attention to the sector,
most farmers are still using their traditional knowledge to solve crop related problems. The
very limited number of specialized experts in the area, coupled with lack of appropriate
technologies, contributed to the present low productive land and low income status of
Ethiopian farmers. Maximizing yield potential and quality need the presence of proper expert
decisions at a field level. In situations where there is a shortage of high level domain
experts, automating crop management decision making has paramount importance. Expert
system technologies can be used for automating crop management decisions as they have
been effectively applied to solve problems in other domain areas of similar nature.
Based on the information gathered at the start of this research, the problem of the country's
agricultural human resource is more intense in the area of vegetable production. With this
background, this research is conducted to develop an expert system model as an attempt to
automate the reasoning strategy of human vegetable experts. There are a number of
approaches to develop expert systems ranging from rule based methods that represent
knowledge in the form of IF-THEN rules to systems that employ machine learning
techniques.
The approach adopted in this research uses the combination of the rule based and neural
network methods with an aim to exploit the best features of the two methods. The system is
modeled to have hybrid architecture by integrating rule based and neural network modules
as a component of one single system.
In the course of building the hybrid model, knowledge acquisition, data preprocessing, rule
generation, knowledge representation and model integration tasks had been performed. In
the rule based module of the hybrid model, knowledge of vegetable experts was represented
as rules. To build the neural network module and perform the integration with the rule based
module, the fast artificial neural network libraries written in the C language were used after
compiling and importing them in the prolog environment. The neural network module is built
to handle user requests that may go beyond the capability of the rule based module. The
artificial neural network module was integrated with the rule based module to create the
hybrid vegetable expert system.
To measure the effect on performance after integration, ten random queries of consultation
requests were presented to both the rule based module and the hybrid system. The hybrid
system responded to eight of them while the rule based module alone provided answers to
only five of these questions. The performance gain observed in the hybrid system is due to
the neural network module embedded in it. The result obtained in this work showed that
integration of the two approaches into one system produced better result and it is
encouraging to advance the system into fully functional vegetable expert system.
Description
Keywords
Information Science