Machine Learning for a layman

The craze about Machine Learning (ML) is at an all-time high, What is it about? Why are people actually interested in this? Let’s see if I can shed some light on ML for a general person.

“Learning from experience”, this phrase is applicable to every person in the world. What if a computer program learns from the experience?

This is basically what Machine Learning is in a very general sense. Here’s a more apt definition for ML from Wikipedia, ‘Machine learning (ML) is the study of computer algorithms that improve automatically through experience and by the use of data’.

The experience a computer can learn from is through the medium of data. Now that we’ve got a general understanding of ML, let’s look at some ideas that can actually make use of this concept.

Supervised, Unsupervised, and Reinforcement Learning

These are the 3 most important concepts ML can be split into. Let’s look at them one by one,

1. Supervised Learning

This subcategory of ML is all about training algorithms to correctly predict outcomes when given input variables ( a.k.a features ). The outcome to be predicted is also called the dependent variable ( a.k.a target ).

Say we want to predict the price of a house given some features that influence the price. Let these features be the Location ( environment ), Size, Age, Renovations, etc. Given all these features and their corresponding price, the ML algorithm can detect patterns and accordingly predict prices for new houses.

The process flow mentioned above isn’t exactly accurate but it’s enough to give us an intuition on how supervised learning works.

As for what Regression and Classification are, it’s pretty simple. If the target variable you’re predicting is a continuous variable then it’s a Regression task, on the other hand, if the target variable is a categorical variable then it’s a Classification task. Some examples,

Regression: Prediction of house price, Revenue of a company, Demand forecasting, etc.

Classification: Is the mail spam or ham, Republican or Democrat, Whether the person defaults in paying back their loan or not, etc.

2. Unsupervised Learning

Here the ML algorithm works with data that doesn’t have a target variable. So, the next obvious question is what can it do with this sort of data?

Unsupervised learning segregates data with similar characterstics/patterns. This is known as clustering.

Examples,

DNA clustering, Customer segmentation, Anomaly detection etc.

3. Reinforcement Learning

This is an approach where in an agent takes action acoording to the environment conditions, but it keeps adjusting the actions in order to maximize it’s cumulative reward.

In simple terms, the computer program faces complex situations and the decision it takes is based on the reward it gets, and trying to get the maximum cumulative reward is the goal.

Examples,

Robotics, Games, Traffic Light Control etc.

Finally let’s look at a general blueprint of the steps taken in a Supervised Learning Algorithm.

Steps taken in a Supervised Learning Algorithm