• Journal of the Korean Data and Information Science Society
• /
• v.28 no.3
• /
• pp.515-524
• /
• 2017

In this study, a measure of discrepancy based on MV (margin of victory) has been suggested that might be useful in determining the size of random forest for classification. Here MV is a scaled difference in the votes, at infinite random forest, of two most popular classes of current random forest. More specifically, max(-MV,0) is proposed as a reasonable measure of discrepancy by noting that negative MV values mean a discrepancy in two most popular classes between the current and infinite random forests. We propose an appropriate diagnostic statistic based on this measure that might be useful for the determination of random forest size, and then we derive its asymptotic distribution. Finally, a simulation study has been conducted to compare the performances, in finite samples, between this proposed statistic and other recently proposed diagnostic statistics.

• Journal of the Korean Data and Information Science Society
• /
• v.27 no.4
• /
• pp.855-863
• /
• 2016

In this study, a simple diagnostic statistic for determining the size of random forest is proposed. This method is based on MV (margin of victory), a scaled difference in the votes at the infinite forest between the first and second most popular categories of the current random forest. We can note that if MV is negative then there is discrepancy between the current and infinite forests. More precisely, our method is based on the proportion of cases that -MV is greater than a fixed small positive number (say, 0.03). We derive an appropriate diagnostic statistic for our method and then calculate the distribution of the statistic. A simulation study is performed to compare our method with a recently proposed diagnostic statistic.

• Journal of the Korean Data and Information Science Society
• /
• v.21 no.2
• /
• pp.371-377
• /
• 2010

In this study, we propose two informal hypothesis tests which may be useful in determining the number of trees in a random forest for use in classification. The first test declares that a case is 'easy' if the hypothesis of the equality of probabilities of two most popular classes is rejected. The second test declares that a case is 'hard' if the hypothesis that the relative difference or the margin of victory between the probabilities of two most popular classes is greater than or equal to some small number, say 0.05, is rejected. We propose to continue generating trees until all (or all but a small fraction) of the training cases are declared easy or hard. The advantage of combining the second test along with the first test is that the number of trees required to stop becomes much smaller than the first test only, where all (or all but a small fraction) of the training cases should be declared easy.