Research On Multi-Classifier Model Based On Adaboost Algorithm And Application On Rainfall Prediction

With the development and progress of social economy,the development and demand of many industries need more efficient ability of the weather forecast.It compared with the traditional pattern of weather forecast,it has more channels to collect meteorological data increasingly,meteorological data has been growth rapidly.It is very important that how to use these amounts of meteorological data effectively for advancing the meteorological fields,meteorological data shows the characteristics of high-dimensional and large-scale,it means that the traditional pattern of weather is hard to research the relation of each dimension of meteorological data,and it is difficult to find the regularity between them effectively,but the data mining technology is relatively mature currently,it has important significance for researching the law of the multi-dimensional attribute in the field of meteorological and seeking out the weather evolution.This paper researched the question of rainfall forecast in the field of meteorological,it had analyzed and grasped the meteorological data mining technology latest,based on this,this paper had proposed that improves the existing weather forecast model.It abandoned the single classifier weather prediction model,and using the integrated strong classifier model achieve the goal,it had studied that the meteorological characteristics of sample data carefully,and had modeled the meteorological data mining and attribute data processing,after this,this paper had research the integrated forecast model deeply,and completed mainly the following work.(1)Aim at solving the problem that current meteorological precipitation prediction model by Adaboost algorithm integrated classifier has low generalization ability and the shortage performance of accuracy,because of learning ability of model has declined,in this article,it had been theoretical derivation proved that the relationship between the integrated classifier's upper bound of error and normalization parameter factor in the process of learning and research the essential way to optimize the Adaboost algorithm,in order to solve the problem of accumulation error,it had adjusted the Adaboost algorithm the way of updating weights in the learning process,improved algorithm had been proposed that it based on the normalized factor and the dependence of the weight correction,the improved algorithm aims to adjust the range of updating the weights of sample in the iteration,the way of adjusting was based on trust,so that,it could achieve the purpose of improving performance,the experimental results showed that this way could improve the performance of prediction.(2)Aim at the problem that precipitation data are always unbalanced in the distribution in the field of meteorological,in this paper,the way to solve the problem of the unbalanced data learning based on the loss cost sensitive,it was the fundamental thought that is improving the Adaboost algorithm,it made the different loss cost sensitive according to different classification prediction results of each sample data,objective function was to minimize the loss cost,this paper had proposed that introduce the punishment factor,the way to update the weights of base classifier had been adjusted in the learning process,and the error of the base classifier threshold by proofing theoretical and deriving formula,The experiment showed that it was well for improving the learning ability of the minority classes in the data set,and it was effective for improving the performance of the integrated classifier.It had been proposed algorithm compared with the traditional data mining algorithm,the results of rainfall experiments and evaluation protocol showed it had good performance.This kind of model enhanced the efficiency of weather prediction,and the meteorological data always reflected the good performance,it can be a better solution for the following meteorological data mining provides.