Towards mental load assessment for high-risk works driven by psychophysiological data: Combining a 1D-CNN model with random forest feature selection.