支持向量机在金融交易分析中的应用研究-智能与分布计算实验室

支持向量机在金融交易分析中的应用研究

姓名	贺江华
论文答辩日期	2008.06.05
论文提交日期	2008.06.10
论文级别	硕士
中文题名	支持向量机在金融交易分析中的应用研究
英文题名	An Application Research on Support Vector Machine in Financial Analysis
导师1	李玉华
导师2
中文关键词	支持向量机;分类;金融交易;距离判别;增量式学习
英文关键词	Support Vector Machine;Classification;Financial Transaction;Distance Discrimination;Incremental Learning
中文文摘	随着经济发展和人们进行商业活动能力的提高，金融交易活动的发生日益频繁，在大量的交易活动中隐藏着许多非正常的交易，洗钱活动也伴随其中，根据专家知识以及相关法律法规人工识别所有可疑交易工作量太大，需要将这些专家知识以及相关法律法规与数据挖掘技术相结合，对大量的交易数据实行自动化、智能化的分析，发现隐藏其中的交易模式，为金融监管部门提供信息指导与犯罪预警，结合金融背景，将支持向量机分类技术运用到金融交易的分析中，通过对历史交易数据的学习可以对未来的交易作出有指导意义的预测。介绍了支持向量机的原理方法，讨论了几类主要的改善支持向量机的方法：二次规划求解方法，分解算法，增量算法以及集成多种技术的分类算法等。接下来对支持向量机算法和神经网络算法进行了比较，说明了支持向量机在学习性能上的特点和优势。金融交易分析是金融监管的重要一环，分析结果对打击防范金融犯罪有重要意义。介绍了金融交易分析的概念，描述了金融交易分析的过程，然后基于支持向量分类给出了一种可操作性强的分析流程，并对分析的各个步骤进行了详细的说明。针对小样本集与大样本集情况分别对已有支持向量机算法进行了改进。小样本集情况下将支持向量机算法与距离判别方法相结合，分别对训练样本集与测试样本集进行分组，通过训练后错分样本的信息指导后续的学习过程。大样本集情况下采用增量式学习方法，解决海量数据处理与训练样本集不完备问题，并根据实际情况融合了类加权思想，对原有训练过程进行扩展。实验表明两种情况下的改进算法都是有效的。
英文文摘	Along with economic development and people's increasing ability to take commercial activities, financial transactions are increasingly frequent. There are many non-normal transactions hidden in large number of trading activities, including money-laundering. There is too much workload to identify all suspicious transactions according to expert knowledge and related laws and regulations. It is necessary to combine expert knowledge and related laws and regulations with data mining technology, then analyze large number of transaction data automatically and intelligently, by which to discover hidden patterns of transactions, provide information guidance and crime warning for the financial supervisory department. In the financial background, SVM classification technology is applied to the analysis of financial transactions to take significative forecast on the future trading through the historical transaction data. First of all, we worked over the theory and method of SVM and discussed several typical methods improved performance of SVM. They are quadratic programming, decomposition algorithm, increment algorithm and integrated algorithm of several advanced methods. The differences of capability between SVM and Neural Network(NN) are reflected. The characteristic and advantage of SVM are also elaborated. Financial transactions analysis is an important part of financial supervision. The result of analysis is significative in the field against financial crime. The concept of financial transactions is introduced, the flow of financial transactions analysis is described, and based on support vector classification method , a analysis flow with strong maneuverability is proposed in detail. The algorithm is improved on small sample set and large sample set respectively. On small sample set SVM algorithm is combined with distance determine method, which can effectively improve the accuracy of the classification of machine learning, so that the traditional method of support vector machine performance is greatly improved. On large sample set incremental learning is taken, and combined with the class weighting thought, by which the training time is shortened and the correct rate also increases.