摘 要 ： 科研经费需求大于投入的现状以及科研资助申请仍面临着低通过率和资助申请过程繁琐复杂的问题。这对有效进行科研项目的申报提出了新要求。传统的评价科研项目申报的方法存在诸如缺乏对历史数据的分析利用、专家知识无积累等缺点，并且科研项目的评价指标与目标之间存在一种复杂的非线性关系，难以用确定的数学关系来描述[9]。数据挖掘中分类技术的出现为解决这些非线性问题提供了技术支持。与传统的评价方法相比，分类预测凭借较强的学习能力为评估是否给予科研资助提供了更好的选择，减少了相关学者在不太可能成功的申请上浪费的时间，有利于相关学者将自身精力投入到能够继续进行的、有发展潜力的科研项目中，对科学进行资助申请具有重要的战略意义和实际意义，同时有利于资助机构进行更加科学有效的决策。本文选取了BP神经网络、C4.5决策树、支持向量机这三种对进行资助申请预测具有可行性的模型，采用墨尔本大学从2005年底到2008年的资助申请记录作为数据对三种模型进行训练。根据识别准确度和ROC曲线对三个模型进行对比评价得出，支持向量机模型更适合对科研资助申请是否会成功的判别，模型的识别准确度到达95%以上。采用支持向量机模型对测试数据进行预测，识别准确度为99.30435%.
Research funding faces the current situation where demand exceeds investment. And grant applications are still faced with low pass rates and complicated grant application process. These put forward new requirements for the effective application of scientific research projects. The traditional method of appraising scientific research projects has the disadvantages such as lack of analysis and accumulation of expert knowledge. And there is a complex nonlinear relationship between the evaluation indicators and objectives of scientific research projects, and it is difficult to describe them with certain mathematical relationships. The classification prediction based on data mining provides technical support for solving these nonlinear problems. Classification prediction is better than traditional evaluation methods in terms of its strong learning ability to judge whether to give research funding. It reduces the time wasted by scholars on applications that are unlikely to be successful, and it is beneficial for relevant scholars to devote their energies to scientific research projects with potential for development and it is of great strategic importance to apply for science funding. Classification prediction can also helps funders make more scientific and effective decisions. This paper selects BP neural network, C4.5 decision tree and support vector machine which are feasible models for predicting grant application. The three models were trained by data from the University of Melbourne funding application from 2005 to 2008 . According to the recognition accuracy and ROC curve, the evaluation of the three models shows that the SVM model is more suitable to judge whether the grant application will be successful, and the accuracy of the model reaches more than 90%. Using the SVM model to predict the test data, the accuracy rate is 99.30435%.
关键词：科研资助申请； 支持向量机； C4.算法； BP神经网络
Keyword: Grant Application； SVM； C4.5 Algorithm；BP neural network
目 录
一、 引言 4
（一） 选题背景与研究意义 4
（三） 本文的研究工作 5
二、 研究方法及措施的选择 6
（一） BP神经网络 科研资助申请的分类预测:http://www.chuibin.com/shuxue/lunwen_205544.html