In some embodiments, the various true positives, false positives, false negative, etc. may be stored and/or presented in a “confusion matrix,” which is a table or matrix data structure that can be used to indicate the classification performance of a predictive model on a set of test data for which the true values are known. The confusion matrix may also be used as a means to compare the accuracy against other predictive models or test the health of a predictive model. FIG. 9 illustrates an exemplary embodiment of a confusion matrix 900. Confusion matrix 900 indicates that a predictive model made 72,000 predictions (n=72,000), which could be, e.g., related to the number of images in an image data set. The image data set may have been tested in a predictive model, such as the 3D convolutional neural network or the ensemble model described herein. The confusion matrix 900 has two predicted classes: “No” (column 902) and “Yes” (column 904), that could, for example, indicate whether a driver behavior was predicted in an image, where “No” could indicate that no driver behavior was predicted and “Yes” could indicate that a driver behavior (e.g., “texting”) was predicted. The confusion matrix 900 also has two actual classes: “No” (row 906) and “Yes” (row 908), that indicate whether the image actually had driver behavior, which could have been determined prior to execution of the predictive model.
