Improving the performance of 1D object classification by using the Electoral College

It has been proven that districted matching schemes (e.g., the US presidential election scheme, also called the Electoral College) are more stable than undistricted matching schemes (e.g., the popular voting scheme for selecting a governor in California), and that the theory can be used in pattern classification applications, such as image classification, where by its nature an object to be classified consists of elements distributed in a bounded 2D space. However, the objects of some pattern classification applications consist of features/values of elements lying on a limited 1D line segment. This paper will prove that districted matching scheme can still outperform undistricted matching scheme in these applications, and the improved performance of districted vote scheme is even more substantial for these 1D objects than for 2D objects. The theoretical result suggests the use of districted matching schemes for pattern recognition of 1D objects. We verified the theoretical analysis through artificial neural network-based approaches for the prediction of start codons of nucleotide sequences. Liang Chen received his Doctor' degree in Computer Science from the Institute of Software, Chinese Academy of Sciences, Beijing, China, in 1994. He is currently Associate Professor and Chair of Computer Science Department, University of Northern British Columbia, Prince George, BC, Canada. His research interests include general artificial intelligence, image processing, bioinformatics, intelligent language tutoring system, computational intelligence, and fast approximate practical algorithms for solving some NP hard problems. Ruoyu Chen is now an undergraduate student (Class 2, Grade 2002) in Computer Science and Technology College at Jilin University, China, majoring Computer Science and Technology. He started to work with Dr. L. Chen in voting theory from 2003. He holds one Chinese Patent. His main research interests include artificial intelligence, natural language tutoring systems, data mining, and networks. Sharmin Nilufar received an MSc degree in Computer Science from Rajshahi University, Bangladesh. She is now a graduate student at University of Northern British Columbia, Prince George, BC, Canada. Her research interests include information retrieval, image retrieval, and bioinformatics.