A Random Forest Concentration Detection Model for Short Video

The detection of concentration is widely used in the fields of education, advertising, and fatigue driving. In order to improve the objectivity, convenience and accuracy of concentration detection, this paper proposes a scheme for using machine learning to detect the concentration of short video. In this scheme, since the eye movement features are used as the basis for the judgment, the detection is convenient. For different types of eye movement data, we use feature selection methods such as logistic regression, wavelet decomposition and approximate entropy, and finally create new combined feature. The combined feature which take into account both eye movement event information and original eye movement information has a strong discrimination. Through research and compare, it is found that after the labeled training based on the questionnaire results, the concentration prediction model established by the combined feature and the random forest algorithm can make the average accuracy of the detection reach 95.52%.