Inductive learning methods for critical decision making applications

机器学习是计算机科学的一个重要研究领域。机器学习的关键活动是从数据集中提取规则，这些规则可用于准确预测未见数据的结果。换句话说，我们试图开发算法和技术，使计算机能够从数据中学习。在这个提案中，我将学习理论方面以及颗粒计算、自组织特征映射（SOFM）和支持向量机（SVM）的技术，并将它们应用于关键的决策问题，如入侵检测、营销分析、财务预测和癌症诊断。作为一项新兴技术，颗粒计算吸引了众多研究者。本研究的目标是增强传统算法的学习能力，并为分类问题和决策应用提供更通用和灵活的粒化算法。在SOFM中，数据是基于一组神经元聚类的，这些神经元具有根据输入向量之间的差异区分自己的能力。当考虑到许多相似度量时，确定哪个神经元最适合特定输入向量的竞争过程是具有挑战性的。将探索博弈论和其他技术，以增强SOFM中的竞争过程，从而使其能够更好地组织和分析更多的数据特征。支持向量机在分类、识别和分类等实际应用中表现出优异的性能。我们将粗糙集理论应用于支持向量机的特征选择问题，并成功地应用于入侵检测问题。对于这个问题，我将继续探索SVM模型可能的改进。我们的最终目标是开发一个决策支持系统，帮助用户确定哪种学习方法最适合不同的应用程序。

Machine learning is an important research area in computer science. The key activity of machine learning is to extract rules from data sets that can be used to predict results for unseen data accurately. In other words, we try to develop algorithms and techniques that allow computers to learn from data. In this proposal, I will study the theoretical aspects as well as the techniques of granular computing, Self-Organizing Feature Maps (SOFM) and Support Vector Machines (SVM), and apply them to critical decision making problems, such as intrusion detection, marketing analysis, financial prediction, and cancer diagnosis.     As an emerging technique, granular computing has attracted many researchers. The goal of this research is to enhance the learning ability of conventional algorithms as well as to provide more general and flexible granulation algorithms for classification problems and decision making applications.     In SOFM, data are clustered based on a set of neurons with the ability to distinguish themselves in terms of differences between input vectors. The competitive process of determining which neuron best fits a particular input vector is challenging when there are many similarity measures are taken into consideration. Game theory and other techniques will be explored to enhance the competitive process in SOFM and thus give it the ability to better organize and analyze more data characteristics.     SVM has demonstrated excellent performance in real-world applications such as categorization, recognition, and classification. We have adapted rough set theory to SVM for feature selection problems and successfully applied it to an intrusion detection problem. I will continue to explore possible improvements in SVM models for this problem.     Our ultimate goal is to develop a decision support system that will help users determine which learning method is most suitable for different applications.