In GIS, data is organized in themes as spatial layers. One of the most significant tasks of GIS is to analysis, spatial layers in order to model spatial phenomena. Since spatial data in GIS are inherently uncertain, a system that handles and infers from such uncertain data is of vital importance. Insufficient consideration to spatial modeling can lead to several problems in spatial decision making, death toll, property damage, financial loss and other hardships.

There is no analytical solution in most spatial modeling. For these modeling, methods inspired by nature sometimes work very efficiently and effectively. These biologically inspired methods are called Soft Computing. The modeling of mineral potential is one of the special cases of the problem of spatial modeling in which no comprehensive model has been developed for it. A mineral potential mapping which depicts the favorability of mineralization occurring over a specified area is an important process for mineral deposit exploration. The purpose of this paper is to suggest several soft computing methods such as a fuzzy inference system (FIS), neural networks and genetic algorithms in a GIS framework for mineral potential mapping.

A typical FIS is composed of two main parts: The Knowledge Base (KB) and the Inference System. The KB composed of Data Base (DB) and Rule Base (RB) stores the available knowledge about in the form of linguistic “IF-THEN” rules. One of the major problems in constructing an FIS is to build the knowledge base. In the conventional design methods, the desired rules and functions are based on the expert's knowledge and experiences. However, we cannot perfectly represent the expert's knowledge by linguistic rules nor choose appropriate membership functions for fuzzy sets. Moreover, converting the experts' knowledge into if-then rules is difficult and often results are incomplete, unnecessary and include conflicting knowledge, since experts cannot express all their knowledge. These problems can be sorted out applying techniques to construct a fuzzy knowledge base of numerical input-output data. In this research, two methods of dividing the input-output spaces and neural network are implemented to deal with this problem.

To evaluate the constructed FIS, the genetic neural network has been applied and its results compared to the results of the FIS. In a genetic neural network, genetic algorithm is used for neural networks in order to optimize the network architecture. As a matter of fact, the topology of the networks is encoded as a chromosome and some genetic operators are applied to find an architecture which fits best the specified task according to some explicit design criteria.

Numerical experimentations showed that the genetic neural network used in this study is the most successful method. It could predict the characteristic of 85% boreholes correctly. The results also indicated that genetic neural network and FIS with RMSE of 4 and 15 respectively are more accurate methods.