Recently, there has been an increasing tendency to use deep learning to quickly predict reliable flow characteristics. Present research is predicting wall normal heat flux using deep learning. As a test, we used stratified turbulent channel flow with Re_tau=180, Pr=0.71 and Ri_tau=0 to make velocity information dominant to temperature information. There is a linear correlation between streamwise wall shear stress and wall normal heat flux but there are some parts where they are not linearly related, too. They closely related to vortex near the wall, or spanwise wall shear stress. Therefore, using several streamwise and spanwise wall shear stress values near a wall normal heat flux value, we designed a model that is a convolutional neural network with 7 hidden layers to predict the heat flux value. We learned DNS data in the model and then obtained approximately 0.97 as the correlation coefficient with DNS data.