The invention discloses a visual fatigue degree detection method based on machine learning. The method comprises the following steps of obtaining the amount of scleral blood of the student's eye through the camera of the terminal after the teaching starts for a predetermined time; using the trained binary classifier to classify the scleral blood volume to determine the credible value of visual fatigue of the students; taking the teaching time as the weight value and weighting the credible value of visual fatigue degree according to the weight value to obtain the true value of visual fatigue degree; if the true value of the visual fatigue degree is not higher than the preset abnormal threshold value, judging that the eyes of the students are not in the visual fatigue state, and if the truevalue of the visual fatigue degree is higher than the preset abnormal threshold value, judging that the eyes of the students are in the visual fatigue state. The method of the invention can detect thevisual fatigue degree of the students in the teaching process.