Confusion-matrix on predictive time-series data?

Hi all, not really a big buff on manipulating machine learning techniques over MATLAB so would appreciate your help!

Essentially, I have a list of timeseries data spanning over time (every 5 minutes-interval). Based on blood glucose data of a patient, I have a working algorithm (in-progress) that predicts a potential time where a meal may have occured. In training the data, I also have access to the actual known meal times. I have provided a screenshot of my seperate arrays (the time series, predicted times, and actual times) below for reference.

Since this is a "continous" data in time, I am really struggling to generate a confusion matrix that can ultimately be used for an ROC curve to optimize my algorithm. Any help is much appreciated!