Modeling Performance of Different Classification Methods: Deviation from the Power Law
msra(2005)
摘要
This project studied the effect of varying the training size for different classification techniques. The learning curves were then regressed using four common equations. In the restricted domain which was studied, the logarithmic equation was the best fit. This contradicts the earlier work carried out on Decision Trees in which the performance was best modeled by the Power Law. The other classification techniques studied in this project were K-Nearest Neighbors, Support Vector Machines, and Artificial Neural Networks, which have not yet been included in such a study. A preliminary study of how the modeling can be used for predicting the performance of the project was also undertaken. The equations which best predicted the performance were not same as the ones which best fit the final curve, and depended on the classification method more than the dataset.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络