I have two classifiers, and I was asked (for learning purposes) to compare their predictions, visually. The target value is a real value (sentiment)
What could be a good way to compare them?
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.Sign up to join this community
If you are predicting a single numerical value for each instance (in which case I would not speak of "classification"), then the best visualization would likely be a scatterplot of predictions vs. actuals. This would allow you to see, e.g., whether you are consistently underpredicting over a certain range. You could also use color to indicate features.
@Stephan Kolassa has provided a good answer. In addition to overlaying scatterplots of predictions vs. actuals, you might also benefit from making a Bland-Altman plot (Tukey mean-difference plot). In that situation, you would not have the actuals represented. Instead, you make a scatterplot of the mean of the two predictions on the x-axis and the difference between them on the y-axis. Such a plot should be horizontal, centered on the $0$-line, and with constant vertical spread. If you wanted, you could test these properties as well, but usually visual inspection is sufficient for people's needs. To some degree, these features can be seen in the scatterplot Stephan mentions, but they may be easier to see in this format. Thus, both plots could make a contribution.