ERPLAB Studio Panels: Plot Confusion Matrices

The Plot Confusion Matrices panel in the Pattern Classification tab is used to plot confusion matrices for decoding accuracy from the selected MVPCsets. The confusion matrices show the likelihood that a true class was classified as each of the alternative classes.

An example confusion matrix is shown below. In this example, participants were shown faces, and the decoder attempted to distinguish between four different face identities. Chance was 0.25 (1/4). Each cell in the confusion matrix shows the likelihood that one of the four faces was classified as each of the four possibilities. For example, when the true label was Face ID 1, there was a 0.4968 probability that it was decoded as Face ID 1, a 0.2045 probability that it was decoded as Face ID 2, a 0.116 probability that it was decoded as Face ID 3, and a 0.1827 probability that it was decoded as Face ID 4.

If the Value to plot is set to Average Confusion Matrix between two latencies, two latency values should be entered into the Latency to plot box, and the routine will compute the average decoding accuracy across the specified latency range and the show those average values in the confusion matrix. If the Value to plot is instead set to Timepoint Confusion Matrix, one or more latency values should be entered into the Latency to plot box, and the routine will plot a separate confusion matrix for each specified latency.

The user can also specify a color palette with the Color to plot menu and specify the minimum and maximum values for the color scale by unchecking the Auto box and setting the limit values.