The fundamental principle of machine learning is to Predict.
Then how will I Predict?
Lets take a very simple example to Predict
- First, we will look in to previous data and find if there is any common pattern and apply that to predict.
- So here the pattern is simple ,it’s Y = 1.5 * X.We can also say Y= m*x ,where m=1.5.
Here Y is called as Dependent Variable and X is Independent Variable
But in real time it won’t be such simple.We will look at such small sample dataset
These are the Profits for some startups in a particular state.We need to Predict the profits for some-other startups
We will Follow the same procedure.
- Look in to previous data where we already know Profits
- Look for any Patterns . But we don’t find any such as it is not straight forward.Here Profit is Dependent on all the other variable.
(R&D Spend,Administration,Marketing Spend,State)-Independent Variables.
So we can assume it as :
Profit = A1*(R&D Spend)+A2*(Administration)+A3*(Marketing Spend)+A4*(State)
There will still some more factors.But will just ignore those.
In Previous example we were easily able to eye ball the sequence.But here it is difficult. So we need some kind of calculator who will help me to find out the pattern!!We will use something called Algorithm who will help me to compute calculations.We will dump the data to algorithm and it will help us to find the pattern i.e.., A1,A2,A3,A4
Next is to validate/Test
Ideally we use 70% data for algorithm and keep 30% data aside for testing.Then we validate and calculate the average error .
Predict the Range
Once the average error is predicted we will predict the profit value+/-
There are lot more factors involved in real time but this is the general concept with prediction.We will look more about this in future blogs.