A Novel Data Selection Method for Improving Supervised Learning Performance of Neural Network

作者：Zhou Yu^*; Su Haibin; Zhang Hongtao

来源：Information-An International Interdisciplinary Journal, 2012, 15(11A): 4513-4518.

• Data selection
• Shadowed sets
• Supervised learning performance
• Core data
• Boundary data

Training data set has a direct and decisive affect on supervised learning performance of neural network (NN). In this work, with the purpose of improving supervised learning performance of NN, we propose a new training data selection approach based on shadowed sets. Taking LVQ model as an example, some experiments are studied on a synthetic data set and IRIS data set to test validity of proposed method. Experimental results indicate that our proposed selection method can keep typical sample data, select informative data, and reduce the number of training data. Using selected data to train NN can effectively improve supervised learning performance of NN.

• 出版日期2012-11