The AdaBoost Algorithm with Prior Probabilities and the Visualization Demonstrated in GIS for Geo-hazard Forecasting
Xianghui Zhao, Zhongliang Fu, Yu Yao, Qing Miao · 2009
The AdaBoost integration learning algorithm is based on the idea of promoting the classification precision through certain combinations by a number of classifiers. This paper puts forward the AdaBoost algorithm with prior probabilities. Each classifier which is used for the combination is usually obtained through the sample collection by certain training. Using the sample to centralize the ratio of different kinds of goals can reflect various classifiers' prior probability. Using this parameter, we can make good use of AdaBoost algorithm to predict hazard quickly and will not cause the phenomenon of over studying. Based on the classification problem of two-classes, experiments with UCI datasets show the validity of the AdaBoost algorithm with prior probabilities. The performance of the AdaBoost algorithm with prior probabilities is better than the traditional AdaBoost algorithm. The AdaBoost algorithm with prior probabilities is confirmed to give better prediction in geo-hazard risk modeling through the visualization demonstrated in GIS.