Abstract: This research work explores soil classification using soil of central Western Nigeria as a case study. This study uses Neural Networks and rule extraction and decision table for the classification of the soils. Soil classification was made more accurate, cheaper and easier with the implementation of Neural Network trained model. A well-developed database capable of generating outputs from a non-linear inputs based on its universal approximation properties was employed, thus, reducing the cost, time, mistakes and labour associated with practical fieldwork. However, the method takes into account the topography and parent material (configuration) of a certain soil sample (location). A generalized Neural Network for soil classification was obtained using the Matlab 6.5 Tool box to build, simulate and test the network. The Feed forward back propagation was used. A GUI (Graphical User Interface) was designed to achieve a high predictive accuracy and enables users most especially a lay user to directly experience an interesting and simulated environment via a friendly user interface of the network. A simplified and explanatory representation of the inputs and outputs was employed. The performance goal was met at 28 epochs using Leveberg Marguartt training algorithm. The neural network achieved an accuracy of 98.20%, which suggested quite a stable network.
E.O. Omidiora , G.O. Oyediran , S.O. Olabiyisi and O.T. Arulogun , 2008. Classification of Soils of Central Western Nigeria Using Neural Network Rule Extraction and Decision Table. Agricultural Journal, 3: 305-312.