Research Initiation Award: An Intelligent Optimization, Clustering and Classification Framework for High Dimensional, Overlapped Classes, and Imbalanced Data

研究启动奖为传统黑人学院和大学的初级和中期职业教师提供支持，他们正在建立新的研究项目或重新指导和重建现有的研究项目。期望该奖项有助于进一步提高教师的研究能力和效率，改善其所在机构的研究和教学，并使本科生参与研究经验。授予哥伦比亚特区大学（UDC）的奖项可能在许多领域产生更广泛的影响。在许多大规模、复杂和网络化的系统中，数据可用性的扩展导致需要提高对从无界大小和不平衡数据中学习的理解，以支持决策过程。针对罕见病、异常行为、微量爆炸物等高度重叠的不平衡课程，开发有效的不平衡学习系统，可以节省资金和生命。从这个项目中获得的知识将有助于改进聚类和分类算法。该项目还将提高UDC本科生的研究经验和培训。提出的研究将开发一种计算成本低的上下文敏感类内聚类方法，通过使用上下文类数据作为边界生成不重叠的子聚类来克服类重叠问题。该算法可以将任意数据分布分离为不重叠的单峰聚类，同时利用中间的上下文数据分布进一步分离聚类。提出了一种新的基于群智能的混合全局优化学习模型，用于同时优化最小二乘支持向量机的特征子集和调谐参数。此外，将创建一种新的粒子群优化自组织算法，通过从有限和未被充分代表的少数类数据中获取有用的知识来提高分类性能。

Research Initiation Awards provide support for junior and mid-career faculty at Historically Black Colleges and Universities who are building new research programs or redirecting and rebuilding existing research programs. It is expected that the award helps to further the faculty member's research capability and effectiveness, improves research and teaching at his home institution, and involves undergraduate students in research experiences. The award to the University of the District of Columbia (UDC) has potential broader impact in a number of areas. The expansion of data availability in many large-scale, complex, and networked systems leads to a need to advance the understanding of learning from unbounded size and imbalanced data to support decision-making processes. An effective imbalanced learning system developed for the highly overlapped imbalanced classes involving rare diseases, abnormal behavior, or even trace explosive, can save money and human life. The knowledge developed from this project will contribute to improved clustering and classification algorithms. This project will also enhance the research experience and training of undergraduate students at UDC.The proposed research will develop a computationally cheap context-sensitive intra-class clustering approach to overcome class overlapping problems by generating non-overlapping sub-clusters using context class data as a boundary. This novel algorithm can separate an arbitrary data distribution into non-overlapping unimodal clusters, while utilizing intervening context data distributions to further separate the clusters. A new swarm intelligence-based hybrid global optimization learning model will be developed to simultaneously optimize the feature subset and the tuning parameters of the least square support vector machine. Moreover, a novel particle swarm optimization self-organizing algorithm will be created to improve the classification performance through obtaining useful knowledge from the limited and underrepresented minority class data.