Skip to main content
. 2010 Jul-Aug;17(4):446–453. doi: 10.1136/jamia.2010.004325

Figure 1.

Figure 1

Histogram summarizing the work saved over sampling results for factorized complement naïve Bayes (FCNB)/weight engineering (WE), FCNB and voting perceptron (VP). The x-axis shows the discretized differences between the methods (FCNB – VP, FCNB/WE – FCNB, FCNB/WE – VP) , and the y-axis shows for how many topics (drug reviews) the given difference in performance occurs. Looking at the white bars, we observe that most of the topics are to the right of 0 on the x-axis, visualizing the advantage of FCNB over VP. Looking at the light grey bars, we observe that, except for two reviews, they are to the right of 0, meaning that FCNB/WE performs better than VP on 13 out of 15 drug groups. Looking at the dark grey bars, we observe that all the bars are in the right half of the interval, visualizing the clear advantage of weight management when using FCNB.