A model is over-fitting when you see the epoch to become 100% accurate or more. to deal with it you can simplify the model by reducing the parameters, for example reduce the convolution neural networks or dropouts. A model is under-fitting when you see the losses do not reduce. Deal it with feeding the model with more data, do back propagation, switching to a non linear model from a linear one, adding hidden layers, convolution neural networks, and random dropouts.