How to optimise the classification model
And what is regularisation and how to implement it

There are different ways and each one is specific to the problem.

But some common ones are:

  1. More Data: More variety and more volume will give better results.
  2. Feature Engineering : (single most important thing) Starting from cleaning the data to creating new features that are relevant, transforming the data into something that is specific to intuition of the domain knowledge. Carefully selected features to get all relevant data related to the problem.
  3. Hyper-parameter tuning: Tuning the parameters of the algorithm
  4. Ensemble Methods : (Probably the easiest and most interesting) Combining of multiple weak models to make a strong model with better prediction by compensating for each other losses.

These are common to most algorithms, but there are many techniques for deep learning other than those mentioned above. Each depends upon the particular algorithm used and the domain and problem to be solved.