Accuracy on Semiboost Are Noticeably

¶ … accuracy on SemiBoost are noticeably higher than any other method with more than one value being above 96 including two values over 97 and a third that was over 98. There was a range of about 2 for the accuracy overall on SemiBoost.

SemTra has nothing above 96 while PGM has nothing above 95. BGCM and ORTSC are all 92's and 93's. As for standard deviations, Semi-Boost ranges from 0 to 2 on the SD range while SemTra does much the same thing except for a few values above 2.

PGM ranges from 2 to 5 as does BGCM with the largest BGCM value being a smidge over five standard deviations. ORTSC ranges from 3 to 6 standard deviations. In short, as the table progresses from left to right, the accuracy values go down and the standard deviations tend to go up.

The standard deviations rise in number as one goes from left to right in the table but the actual width of the range of the values varies little overall, almost always being roughly 2 standard deviations for each set.

In a similar way, the accuracy figures do the opposite as they tend to go down as one goes from left to right. The Semi-Boost figures are all 95+ but none of the ORTSC figures are above 94.

There is a noticeable drop from SemiBoost to SemTra and then it gets more subtle from column to column in terms of the accuracy values.

Multi-Class Problem

The values for accuracy on the multi-class problem are much more disparate. Indeed, they are all over the place.

SemiBoost has values in the low forties except for a single value in the high 30's.

SemTra has values in the 93 to 94 range while PGM has values in the low to mid-80's.

BGCM has accuracy values in the mid-50's and ORTSC is in the mid-40's.

While the accuracy values vary widely from method to method, there is not a ton of variation between each value within each method.

Most of the accuracy ranges have an overall range of 2-3 points at most.

The standard deviations are a lot more disparate, especially given what a standard deviation is at the end of the day.