How to decrease loss and increase accuracy in a random data experiment?

Question

I just step into the world of Deep learning and made some little experiment to learn how to adjust the model to decrease loss and increase accuracy, No matter how many methods