Reducing the noise

Today I learnt about signal to noise ratio. We use PCA to maximum the ratio for better model performance.

We apply various techniques first to prepare the data. Get z-score, get coefficient of variation and so on.

As always, trying to contextualise it in real life terms, I am thinking of it this way. Our mind cannot processes a ton of information at the same time very effectively to decide if a lot of it is redundant. We need good quality information or try to peel the layers of information we get when dealing with our work or relations, then infer something, then think again, then infer something better and then decide. We are reducing noise in the data we get as we spend more and more time in our decision making progress.

Models are the same, if what it essentially needs is thumb print and we keep giving entire hand’s information - it would not perform well. 

Data science and statistics beautifully plays with numbers with the help of algorithms, formulas and so on to get to the right scale and level of information that would make the decision making easier.