Abstract: Aiming at the problem that the decision tree model of small sample training data is difficult to obtain a high prediction accuracy rate on multi classification problems， this paper establishes a breakout prediction model based on one vs rest genetic algorithm to optimize the decision tree. By making full use of the global search ability and robustness of genetic algorithm， strengthening the search control and supervision of the optimization process， the accuracy of the model is improved.Combined with the continuous casting production data of a steel plant， the breakout prediction model of genetic algorithm optimization decision tree based on one kind of congruence method was tested. The tests shows that genetic algorithms can achieve an accuracy of 98.39% and a reporting rate of 100% for the optimization decision tree steel leakage prediction model based on a class to class genetic algorithm after only 10 iterations. Compared with traditional decision tree algorithms， this algorithm can achieve higher accuracy and better generalization in very few iterations.

Key words: Genetic Algorithm, Decision Tree, Continuous Casting, Steel Lakage Prediction

摘要： 针对决策树模型在小样本训练数据的多分类问题上难以获得较高预报准确率的问题，建立了一种基于一类对余类法的遗传算法优化决策树漏钢预报模型，通过充分利用遗传算法的全局搜索能力以及其具有的鲁棒性，加强搜索控制和优化过程的监督，提高了模型的准确性。结合某钢厂连铸生产数据，对基于一类对余类法的遗传算法优化决策树漏钢预报模型进行了测试。测试表明，遗传算法仅在10次迭代后就可以使基于一类对余类法的遗传算法优化决策树漏钢预报模型达到98.39%的准确率和100%的报出率，该算法相比传统决策树算法可以在极少的迭代次数下得到更高的准确率和更好的泛化性。

关键词: 遗传算法, 决策树, 连铸, 漏钢预报