Want to make sense of your data?

Book FREE Machine Learning call.

Book a Free Demo
Book a Call
Book a Demo

Automate Your Data Work

Intersect enables you to build data workflows and turn them into easy-to-use interfaces.
Get Free Demo

How Does Machine Learning Work | Simple Examples

An epic tale of the search for the best mango can help you understand what machine learning is all about.
Ankit Gordhandas
Ankit Gordhandas
May 7, 2020
Updated on:
October 16, 2020

Automate Your Data Work

Intersect enables you to build data workflows and turn them into easy-to-use interfaces.
Get Free Demo


Download The Book

Ever wondered how Netflix predicts whether a show is a good “match” for you? Or how Gmail classifies an email as spam or not spam? Or how Amazon can recommend your next purchases? The technology powering all of that is machine learning. Recently the phrase machine learning has become an all-too-familiar buzzword. But what is machine learning?

Understanding Machine Learning

Computers are like big calculators. They have big computational power that makes it very easy to calculate millions of data in a fraction of a second. Most of the apps on your computer do some kind of arithmetic calculation in the back and nothing more. Actually, even streaming your favorite tv show on Netflix, is probably calculated in a couple of simple operations.

For us humans, these computational techniques come harder, but we can always learn how to do it, by following a couple of instructions. What comes naturally for us and extremely hard for the computers, is learning from past experience. For example, finding a face in a crowd of people or making a difference between a dog and a cat.

For this kind of learning to happen, we think that it requires some kind of intelligence, that the computers lack. We want to teach our machines to “learn” from past experiences, and utilize the fact that they are fast, don’t get hungry or tired, and can store multiple information at once.

For that, we build machine learning algorithms or “recipes” to guide the computer to learn from previous data, and solve these “harder” cognitive problems for us.

Let's look at 2 examples to understand the concept even more.

Example 1: Identifying the Perfect Mango

A wide array of mangoes

To understand the concept even more, let’s start with a simple example: let’s say you enjoy eating mangoes (who doesn’t?!). Picking the perfectly sweet mangoes can be a challenge though. So you decide to embark upon an experiment. You buy some mangoes from the store -- some large and some small. You then note down your observations in your diary:

Based on these observations, you conclude that you should always buy larger mangoes to maximize your chances of eating sweet mangoes.

Now that you have cracked the code, you are excited to forever live your life with sweet mangoes. The next time you go to the store, you only buy larger mangoes. But alas! Some of your mangoes end up being tart. Luckily, you have noticed another pattern: you have realized that of the larger mangoes, the sweeter mangoes tended to have a bright yellow skin (as opposed to pale yellow or yellow interspersed with green). So you adjust your findings: “Large + yellow mangoes = sweet mangoes”. 

That weekend, your best friend visits you, and knowing your love affair with mangoes, she decides to bring a bunch of mangoes for you. Given you haven’t shared your wisdom with anyone, many of the mangoes she brings are small. Much to your surprise though, some of the smaller ones pack a punch! On closer inspection it turns out that the smaller mangoes that were softer happened to be sweeter, while the small but firm mangoes leaned more on the tart side of the spectrum. Which means, your recipe for finding the perfect mango = “Large + yellow or small + soft”.

What you have done here is learning. You have identified patterns from your observations and turned them into a mental model -- so that the next time you come across a mango, you would just have to mentally observe the size, color and firmness, and translate those into a flavor profile. 

What if you could record similar observations from millions of such mangoes? Surely you could come up with an extremely accurate flavor predictor! But how on earth would you mentally process all those observations to come up with a mental model to output the flavor?

That is where machine learning comes in! You can throw a large number of observations at a computer. The computer then finds statistical patterns to define a model -- so that next time you input a new set of observations, it can predict the desired output. This process by which the computer identifies patterns is machine learning.

Example 2: Predicting house sale prices

Let’s look at a more relevant example. Levi is a real estate investor, and he is hell-bent on outperforming other investors out there, so he gets his hands on all real estate transactions in his county for the past few years, along with details about each house. He figures that the square footage of the house is likely the best predictor of the sale price of the house, so he plots the area and the sale price on an x-y graph. He discovers that the relationship is fairly linear.

The next time a house goes on the market, he uses this relationship to try to make the perfect bid for the house, but unfortunately is way off. He realizes later that the house sold for more than he had predicted because it had more bedrooms than are standard for a house that size. He now makes a fancy 3-D plot to understand the relationship between square footage, number of bedrooms and the sale price of houses.

Once again, he realizes that people also care about the neighborhood. And the quality of construction. And the size of the backyard. And many other things.  Given so many dimensions, it would not have been realistic to easily try and figure out the relationships between all these variables and the sale price of the house.

Had Levi wanted, he could have written some computer code to take in all the real estate data, learn the patterns, and accurately predict the sale price of a house the next time one came on the market -- a program to do machine learning!

What else can machine learning be applied to?

Today you can use machine learning for:

  1. Classify which leads are most likely to purchase your product
  2. Predict the lifetime value of a user on your platform
  3. Understand when your parts will next need maintenance
  4. Recommend upsells

I want to start doing machine learning

If you have been collecting data, and want to make it work for you, you need to start doing machine learning. Maybe you need to identify users most likely to churn? Or forecast revenue for the next few months? Perhaps qualify sales leads?

Intersect Labs makes it super easy to do machine learning. No prior coding experience or mathematical background necessary! All you need is a spreadsheet. Book a call with us to see how you can start using this advanced technology in a few clicks.

Enjoyed The Read?

Don’t miss our next article. Get updated on current data trends and powerful automations that can solve your data challenges.

How we integrate data into our work is just as core as the data itself.

No items found.