What is Machine Learning?
Let’s understand it my way!
Before understanding what Machine Learning is, let us see how we learn.
In my childhood, my parents used to show me different species of mammals, birds, trees and tell me which is what. ( I am using the word “mammals” , not “animals” because bird belongs to animal kingdom ).
I used to grasp the features and predict what is what. After the initial observations, I told myself that those who can fly are “birds”, those who cannot fly, but move from one place to another are “mammals”, and those which cannot move are “trees”.
Now, one day I saw a Bat for the very first time and I asked my dad to name this bird. I was surprised, when I came to know that this is indeed a mammal. Wow! I did not know that mammals too could fly.
Therefore, I updated my understanding and started predicting again.
Machine Learning is similar.
Above mentioned scenario in Machine Learning is called Classification and it is a supervised learning.
Why Classification?
- Because, based on few features (predictors), we are trying to classify the object ( whether it is a mammal or bird or tree).
Why supervised learning?
- Because, my parents were my supervisors and they used to train me showing different species of mammals, birds and trees.
What is a Machine Learning model and what is training and testing data?
- My knowledge base is equivalent to machine learning model, which got trained by the data my parents used to provide me.
Whatever my parents taught me initially , that dataset is called training data in Machine learning.
After that, they used to test whether I was learning properly or not. They showed me different objects and I had to predict whether that is a tree, a bird, or a mammal.
I occasionally made few mistakes especially when it did not fit into my existing knowledge base.
These dataset is called testing data.
What is model accuracy?
- Say, based on my knowledge base, I classified 8 out 10 objects correctly, then my model accuracy would be 80%.
What do they mean by different machine learning algorithms?
- Say, I had my cousin with me, who was also listening to all these conversations between my parents and me. However, he developed his own understanding, and predicted 9/10 correctly. Then he acted as a different machine-learning algorithm and his model accuracy for the same dataset was 90%.
The end goal is to identify the hidden pattern from the dataset, so that machines can learn to automatically predict effortlessly, just like we can do it now. :)
If I catch your interest, let me hear from you. Feel free to comment! :)