Machine Learning is broadly classified into three types, namely Supervised Learning, Unsupervised Learning, and Reinforcement Learning. Reinforcement learning is still new and under rapid development, so let’s just ignore that in this article and deep dive into Supervised and Unsupervised Learning.
Top Machine Learning and AI Courses Online
Before moving into the actual definitions and usages of these two types of learning, let us first get familiar with Machine Learning. Machine learning is an application of artificial intelligence that provides systems with the ability to automatically learn and improve from experience without being explicitly programmed, and this is just the textbook definition of Machine Learning, as this article is mainly written for the newbies of Data Science and Artificial Intelligence field let me make this more clear and interesting for you so that you can understand and interpret it better.
Let us consider a baby as our machine and we need to help the baby learn the different numbers in our number system. In order to help the baby learn we need to show the baby a different number and tell what each number is.
Doing this part repeatedly helps the baby learn and memorize the numbers. This is nothing but the ability to automatically learn and improve from experience without being explicitly programmed i.e. Machine Learning.
Trending Machine Learning Skills
Enrol for the Machine Learning Course from the World’s top Universities. Earn Masters, Executive PGP, or Advanced Certificate Programs to fast-track your career.
1. Supervised Learning
Let us start again with the classic textbook definition of Supervised Learning and make ourselves familiar with the baby example that we earlier took. Supervised learning is the machine learning task of learning a function that maps an input to an output based on example input-output pairs. It infers a function from labeled training data consisting of a set of training examples.
I hate the definitions that are written in any textbook as they are so formal to understand, rather I would prefer a friend to explain the definition in his own thoughts. In the long run whenever I try to recollect a definition, eventually the explanation given by a friend with an example pops up and makes my life easier. So, in this article let me be that friend to you.
Let us again take the baby example we considered earlier, in this case, we need to make the baby learn and identify the different fruits that we have. Let us consider Apple and Orange as our two fruits, and we start by showing these two pictures to the baby. We also tell the baby which picture is which fruit.
Looking at those pictures, the baby learns that the fruit will be round and red colour fruit is Apple, and orange colour fruit is Orange. Now let us show the baby a new picture of an Orange and ask him to find out whether the picture is Apple or Orange.
The baby predicts that the fruit is Orange. The baby correctly predicts the fruit as Orange because we have labelled the two fruits Apple and Orange into two categories and have asked the baby to learn what fruit is what. This is how Supervised Machine Learning works if we replace a machine with a baby.
Supervised Machine Learning is further classified into two types of problems known as Classification and Regression.
From the name itself, we can get to know that this is a Machine Learning problem where we need to classify the given data into two or more classes. The above example that we have taken is a Classification problem as we need to classify the given pictures into either an Apple or Orange class.
When we have only two classes to classify our data, then it is called Binary Classification. But in real-world data, we tend to have more than one class and it is called Multi-Class Classification. These types of learning are used by the majority to identify spam emails, classify customers, check whether a customer Churns from the operator, and many more use cases.
Industry applications of classification
Some of the real-life applications of classification are –
- Image classification
- Fraud detection
- Speech recognition
- Spam filtering
- Document classification
- Facial recognition
- Medical diagnostic tests
- Malware classification
- Product categorisation
- Customer behaviour prediction
Types of Classification Tasks
Some of the classification tasks in machine learning are mentioned below-
- Predictive Modelling
Classification is a problem of predictive modelling where the class label stands to be in anticipation.
- Binary classification
As the word binary suggests, in classification binary is a type of problem that has only two possible outcomes. For example, (yes or no), (true or false), (spam or not spam), and so on.
- Multi-class classification
Opposed to binary classification, multi-class classification is a type of problem that can have more than two possible outcomes. In this case, each problem is assigned to only one label. For example, classifying images, classifying species, and so on.
- Multi-label classification
It is a type of problem that may have more than one assigned class label to the data. The model would have multiple outcomes in this scenario. For example, an image can have multiple objects.
- Imbalanced classification
In the presence of an unequal distribution of data, an imbalance is created. Imbalance classification refers to the classification method where the data distribution is skewed or biased.
Also Read: Career in Machine Learning
Regression on the other hand, deals with continuous data, such as predicting your salary based on experience. In this case, we do not need to put the data into any classes but need to predict the continuous value based on the continuous data we have.
These types of problems have continuous columns in their data set, whereas Classification tends to have categorical columns. These types of learning are used to predict the financial growth in the next quarter for any company, student marks based on his previous marks, and many more.
Industry applications of Regression
- Comparing with competition
- Capital Asset Pricing Model (CAPM)
- Problem identification
- Better decision making
Types of regression analysis
Some of the different types of regression analysis are mentioned below –
- Simple linear regression
It is a relationship between a dependent variable and an independent variable. The simple linear regression model reveals a liner or a slanted straight line.
The model has an expression, as depicted below;
- Y= dependent variable
- X= independent variable
- a= intercept
- b= slope
- ϵ= residual
- Multiple linear regression
It is a statistical process that helps in using multiple explanatory factors to predict the outcome. Multiple linear regression is a method to represent a relationship between dependent and independent variables.
The mathematical representation for MLR is;
y=ß0+ ß1 x1+ …………..ßn xn + ϵ
- y = the dependent variable’s predicted value
- B0 = the y-intercept
- B1X1= B1 is the coefficient for regression of the first independent variable X1
- … = Repeat for as many independent variables as you’re testing.
- BnXn = the last independent variable’s regression coefficient
- ϵ = model error
- Non-linear regression
Data are fitted to a model and then numerically displayed. The non-linear regression connects two variables (X and Y) in a curved (non-linear) shape.
The model aims to minimise the sum of squares as the sum of squares is a statistic which helps to track how much Y observations differ from the non-linear function which was used to anticipate Y.
4. Unsupervised Learning
Unsupervised learning is a type of machine learning that looks for previously undetected patterns in a data set with no pre-existing labels and with a minimum of human supervision.
In contrast to supervised learning which usually makes use of human-labeled data, unsupervised learning, also known as self-organization, allows for the modeling of probability densities over inputs. Let us consider the baby example to understand Unsupervised Machine Learning better.
Let us use a group of cats’ and dogs’ pictures as input in this example, in earlier examples the baby knows that the pictures are of Apple and Orange as we have labelled and categorized them.
In this case, the baby doesn’t know anything and hence cannot categorize which one is a cat and which one is a dog. But can tell that few of the pictures look similar when compared to the other few. In this case, we cannot label the data, but we can still find patterns in the data. This is how Unsupervised Machine Learning works.
Applications of unsupervised learning
Some of the real-life application of unsupervised learning is-
- Customer segmentation
- Understanding of different customer groups
- Clustering DNA patterns
- Anomaly detection
- Recommender systems
The above-taken example clearly describes the Clustering problem, and we need to cluster our dataset based on the patterns that we find in our data. Clustering is a very important Machine Learning problem, and many companies tend to use this technique to find valuable patterns and insights from their data.
Popular AI and ML Blogs & Free Courses
Examples of clustering
The supervised and unsupervised learning examples differ. The industry examples for unsupervised learning are mentioned below-
- Anomaly detection
Any type of outliers can be detected with the help of clustering. Organisations with invested efforts in transportation and logistics may use anomaly detection to identify the logistical obstacles.
- Customer and market segmentation
Clustering can help the users to group people having similar traits and create customer personas. This yields results in effective targeting and marketing.
Types of clustering
There are various types of clustering which can be used in different ways-
- Exclusive clustering- One piece of data belongs to only one cluster.
- Overlapping clustering- The data items can be a member of more than one cluster.
- Hierarchical clustering- Helps to create a hierarchy of the cluster items.
There are various clustering algorithms that are at play-
It is used for exclusive clustering. It helps in putting the data into the predefined numbers or clusters known as K. Items get assigned to the nearest cluster centre called centroids.
- Fuzzy K- means
It is an extended part of the K- means algorithm. Fuzzy K- means denotes that the data points can belong to more than one cluster with a certain level of closeness with one another.
- Gaussian Mixture Models (GMM)
It is used in probabilistic clustering because of the unknown mean or variance. The model assumes that there is a certain number of Gaussian distributions. The algorithm helps to decide which cluster the data belongs to.
So these were the major difference between supervised and unsupervised learning. Now let’s summarise the differences in the form of a table;
|Supervised vs Unsupervised learning
|The users know what they are looking for in the data
|The users do not know what they are looking for in the data
|Classification and regression problems
|Clustering and association problems
|Result in accuracy
|Less accurate results
In this article, we got to know about the different types of Machine Learning, got to understand those taking an easy-to-understand example, investigated the further divisions of each learning. This article covers only the basics of Machine Learning problems, each type of problem has different types of Machine Learning Algorithms.
If you’re interested to learn more about machine learning, check out IIIT-B & upGrad’s PG Diploma in Machine Learning & AI which is designed for working professionals and offers 450+ hours of rigorous training, 30+ case studies & assignments, IIIT-B Alumni status, 5+ practical hands-on capstone projects & job assistance with top firms.
What is an example of unsupervised learning?
The examples of unsupervised learning are mentioned below - K means, KNN, Anomaly detection, Hierarchical clustering, Apriori algorithm, and Neural networks.
What are the two types of supervised learning?
Some of the types of supervised learning are Regression, Naive Bayes, Classification, and Random forest
What are the two main types of unsupervised learning?
The two main types of unsupervised learning are Clustering - It helps in grouping the unlabelled data based on their differences and similarities. Association rules - It helps to find a relationship between the points in the dataset.
What is an example of supervised learning?
Some examples of supervised learning are Prediction of house prices, Classification between two different items like cats and dogs, and Weather forecasting