Can Computers Learn Like Humans?

Have you ever wondered how streaming services providers like Netflix and Amazon Prime seem to get better and better in recommending you the next video to watch? Or how can banks accurately identify the customers who are likely to default even before offering them a loan? In short, how are the computers becoming so smart?

The answer lies in the rapidly evolving field of “Machine Learning”.

But what is Machine Learning?

According to Wikipedia, Machine Learning is defined as the study and development of computer algorithms that improve automatically through experience, and is a subset of Artificial Intelligence (AI). In simpler terms, Machine Learning strives to teach computers (as you would do with toddlers) to take decisions by feeding them with data.

Machine learning (ML) is broadly classified into two categories:

a.    Supervised Machine Learning

b.    Unsupervised Machine Learning

Let’s dive deeper into these categories to understand how they can help organisations achieve their business goals.

Supervised Machine Learning

In supervised Machine Learning, we have one or more independent variables (Xi) and a dependent variable or the variable that we want to predict (Y). A mapping function (f) is derived from the independent variables to the dependent variable with the help of an ML algorithm. Mathematically, it can be represented as:

Y = f (Xi)

Where,

Y = Dependent Variable

Xi = Independent Variable(s)

f = Mapping Function

It is called “Supervised” because the algorithm learns from a well-labelled data-set. By “well-labelled”, meaning that the interpretation of the data is provided to the algorithm (e.g., a picture of the car is labelled as such), so this process can be thought of as having a teacher closely supervise the learning process and hence, the name.

Supervised machine learning can be further grouped into two broad categories:

1. Classification problems
2. Regression problems

Classification Problems

In a classification problem, the variable to be predicted takes the form of distinct categories. A few examples of classification problems:

1.    Detecting Fraud

In case of fraud detection, there are usually two categories, which are “Fraud” and “No fraud”. These two categories can be represented by 0 and 1 and a supervised Machine Learning algorithm can be used to predict the correct category for each transaction.

2.    Detecting a Disease

In the healthcare sector, a possible use-case is disease detection, i.e., whether a person has a particular disease based on their blood tests or X-Rays. Therefore the categories can be “Disease” or “No Disease”, which can again be represented by “Yes” and “No” or 0 and 1.

The image below is a simple illustration of categorisation / classification problem.

The algorithms can also be trained to predict multiple categories, for example:

1.    Disease Prediction

This problem involves predicting the disease a person can suffer from based on their genetic make-up and lifestyle factors. In this case, there can be multiple categories of diseases to choose from, like diabetes, heart disease, mental disorder, cancer, etc.

2.    Predicting the Colour Range of Diamonds

The colours of diamonds can depend on their size and the cutting-angle. Classification methods can be used to predict the diamond’s colour with a wide range of potential outcomes, e.g., colourless, near colourless, faint yellow, very light yellow, light yellow.

Regression Problem

In a regression problem, the variable to be predicted (the output variable) takes the form of integer or real numbers. Some typical use-cases of regression are:

1.    Share price prediction, where share prices are in real numbers of dollars.

2.    Predicting the weight of a person, which again is a real value and can be expressed in kilograms.

3.    House price prediction, where the predicted price is a real number. An example data table that can be used to train the machine learning algorithm is below.

Some of the popular supervised Machine Learning algorithms are:

1.    Linear Regression for regression problems.

2.    Logistic regression for classification problems.

3.    Random forest, decision tree, support vector machine (SVM), and K-Nearest neighbours (KNN) for both classification and regression problems.

Unsupervised Machine Learning

As opposed to supervised Machine Learning, the unsupervised variety does not use any labelled data or defined output variables. The goal of unsupervised Machine Learning is to model the distribution of the data using the combination of the input variables and Machine Learning algorithms. The algorithms then discover the underlying structure of the data all by themselves.

Unsupervised Machine Learning problems can be broadly classified into three categories:

1. Clustering
2. Association
3. Recommendation engines

Clustering

Clustering is used to discover the different groups inherent in the data. The clustering problem can be better understood with the help of an example.

If we want to group customers based on their purchase behaviour, we can use clustering algorithms. These algorithms analyse the various input variables (in this case, customer transactions) and form clusters of data-points with similar properties. After the formation of the clusters, the data scientists can name the clusters by analysing the properties of each cluster (e.g., low-income, middle-class, affluent).

This process is depicted in the image below.

Association

Association is used to discover rules that describe the data by finding associations and relationships among them.

Association rules are widely used in Market Basket Analysis, which is a technique used to identify the relationship between different products or items.

These rules are widely used by e-commerce companies to predict the association between different products and then recommend customers items that are frequently bought together, because it could encourage customers to buy more items, leading to an increase in sales.

You can observe the outcomes of these algorithms on popular e-commerce websites, such as Amazon or E-Bay, which show products under “Frequently Bought Together” or “Customers who bought this, also bought X” (with a list of items)

Recommendation Engines

Recommendation engines are systems that help recommend products and services to customers by analysing customer data and learning about their preferences using machine learning algorithms.

Recommendation engines are frequently used by streaming service providers like Amazon Prime and Netflix to recommend movies and TV series to their users by analysing past transactions and inferring the genres the user likes.

They are also used by e-commerce websites to learn about user preferences and recommend a wide range of products under the section “Products You May Like.”

Recommendation engines, therefore, help in providing a customised user experience, leading to satisfied customers, while helping the organisations boost their revenues.

The popular unsupervised Machine Learning algorithms are:

1. K-Means algorithm for clustering.
2. Apriori algorithm for association problems.
3. Collaborative Filtering for recommendation engines.

Key Takeaways

1. Machine Learning helps make computers smarter by giving them the ability to learn and “think” like humans.
2. Machine Learning can be split into two main types, supervised and unsupervised.
3. The key difference between supervised and unsupervised Machine Learning is that the data required for the supervised learning process is well-labelled, while the data for the unsupervised learning process is un-labelled.

---

Could Machine Learning help your business achieve its goals? At MindGap, we can help you figure it out. Get in touch for a free consultation.