classification algorithm

Land use identification and determination of their spreading level in the region is one of important factors in the natural resources and environmental studies. Preparing land use map from satellite data is one of the fastest and most cost-effective methods. This study’s aim is to determine the best algorithm for TM satellite images classification between 6 supervised classification methods including maximum likelihood, mahalanovis distance, minimum distance, parallel pipe, support vector machines and binary codes to extract land use map of Zenouz Chai watershed.For grassland, wasteland and abandoned and agricultural land use, 30 training samples and for badland land use 45 training samples were prepared separately using ground control points. Results were assessed for satellite image classified using by overall accuracy index, kappa coefficient, producer accuracy and user accuracy. Investigation of TM image classification accuracies showed that the maximum likelihood algorithm with overall accuracy coefficient (84/73 %) and kappa coefficient 65/0, have the higher efficiency for image classification into four land use classes. The results showed that maximum likelihood method has more capabilities than other methods for land use map preparing. Therefore, the results of this study can be used to provide land use maps with higher accuracy using maximum likelihood method to assessing the environment and natural resources works in the areas with the same situations.

Fault is one of the main processes in tectonics which has relationship with phenomena such as earthquake. Therefore, awareness of the effect of environmental factors on fault occurrence and recognition of high risk areas is very important, that these goals in the present study have been investigated. The research area is Qara-Qum where after determining its faults, altitude, slope, direction, climate, land use, geomorphology, geology, erosion, precipitation, flood zones, pedology, stream power, topographic ruggedness index, distance from road, waterway, anticline and syncline were extracted. Then, data mining algorithms including of decision tree, random forest, cumulative, backing machine, logistic regression and neural network in R software are used to identify the value of variables and bivariate statistical methods including of information value and area density for identification of the values for each variable class fitted in fault occurrence. The accuracy of classification algorithms with ROC curve showed that based on input variables, random forest and support vector machine algorithms with 88% and 86% area under a curve had the best performance in classifying fault occurrence, respectively. Finally, according to the Gini coefficients in random forest algorithm, the zoning maps obtained by combining this algorithm were prepared and validated by bivariate statistical methods. According to this algorithm, the height, pedology and topographic ruggedness index variables, identified as the most important parameters in the fault occurrence respectively. Based on the zoning maps evaluation, information value and area density methods, around 52 and 35 percent of faults placed in very high risk class respectively. Therefore, the information value method was more accurate in identifying fault-sensitive zones. According to the results, data mining methods were introduced as a useful tool in fault risk management. It is also necessary to pay attention to environmental variables, especially topography, during the basin management and land use change stages.