4 Different Types of Machine Learning

Machine learning has certainly become a buzzword these days, spreading its wings in various industries, and transforming the way we look at the world. Even today, there are many use cases, where many of us are using or experiencing Machine learning knowingly or unknowingly.

In this article, we are going to discuss some basic types of Machine learning. If you are a Machine learning enthusiast or are just curious about the types of machine learning, as you go on reading this article through the end, you should be well familiar with the different types of machine learning.

4 Different Types of Machine Learning

Note that in this article, we are going to classify Machine learning, on the basis of the type of learning, or we can say on the basis of the level of supervision. Here is how we classify Machine learning, on the basis of the level of supervision –

• Supervised Learning.
• Unsupervised Learning.
• Semi-Supervised Learning.
• Reinforcement Learning.

Now, that we have listed these different types of Machine learning, on the basis of the level of supervision, let’s discuss these different types in brief so that you can be comfortable understanding the concept. In this article, we are using a general language, which should be easy to understand for a wider range of people, so, this may not involve too much technicality, but we will try to explain the concepts as simply as possible, with help of some examples, which would make it easier to relate.

Machine Learning Types with Examples

1. Supervised Learning

Let’s first understand supervised learning with help of an example, before we move towards the quite formal definition of Supervised learning. Let’s say that there is a small kid, who does not know what is a dog, and what is not a dog. So, we show it tons of images, and on each image, we tell it whether it is a dog or not a dog. For example, let’s say that you show image number one, which is of a dog, and you say that it is a dog, and on another image, it is not a dog, so you say it is not a dog.

You did the same thing around a million times (for example), and then you just ask the baby, showing some another image, whether or not it is a dog, and you would find that the baby would be able to distinguish between a dog, and not a dog.

So here, if we try to get what has happened in this process, we had a labeled dataset, which means that we were showing pictures, with the labels that is it a dog or not. So, we can say that the training is done on the labeled dataset.

In short, we can say that in supervised learning, the algorithm is trained using a labeled dataset. Then this information is used by it, to predict the output. This thing is done under supervision, which is why it is called Supervised Learning. We can say that inputs are mapped to the outputs.

Supervised learning can also be classified into two types of problems – Classification and Regression problems. Let’s try to discuss them in brief as well.

Classification Problems in Machine Learning

In case when we are having the output variable in the form of categorical values, the problem would be the classification problem. For example, if you want to classify the mail as spam or not spam, or you want to predict if the person will purchase some product on the basis of given data, or you just want to predict whether it is a cat or dog, based on the image.

Classification can also be classified as binary classification or multi–class classification. If you are having only two classes for classification, then it is said to be binary classification, and on the other hand, if you are having more than 2 classes, it is said to be multi–class classification.

Here are some of the classification algorithms, which are more frequently used –

• K nearest neighbor
• Naive Bayes
• Logistic Regression
• Support Vector Machines
• Decision Tree
• Random Forest

Regression in Machine Learning

As in classification, usually, the output data is categorical, and on the other hand, if the output data is numerical like if we were to predict the price of the house, or the price of some stock, or some probability, we treat the problem as a regression problem. We can say that the value here is continuous rather than discrete.

Here are some of the Regression algorithms which are more frequently used –

• Linear Regression
• Polynomial Regression
• Lasso Regression
• Decision Tree regression

So, this was about supervised learning. Just to summarize, we can say that in the case of supervised learning, we are working on the labeled data. At times, you need to perform supervised learning. It is very simple to understand and implement, and you can freely explore more about the concept of Supervised learning, but we hope that you got a broad picture of Supervised learning by reading this!

2. Unsupervised Learning

As the name suggests, here, there is going to be no supervision involved. Here, we are using unlabeled datasets for training the machines. The model has to learn from the data, find some patterns in the data, and categorize, or cluster the data accordingly.

Using this type of learning, one is able to discover the hidden patterns in the data and trends in the data.

For example, if you provide a lot of pictures of cats and dogs, which are basically unlabeled, the machine would be able to cluster similar things together, like it would help us get all the dog images together, and all the cat images together, though it may not know what is what. Once the machine establishes some patterns or trends in the data, it can simply do this without supervision or human interference.

We can classify Unsupervised Learning in some of the following ways –

• Clustering
• Association
• Anomaly Detection

Let’s try to discuss these things in super brief so that we can just have an idea about all this stuff.

Clustering in Machine Learning

If you simply google the term “Cluster”, you would get a big picture from the name as well. In this technique, things are grouped on the basis of the similarities in the features. This involves dividing the data points into a number of groups, such that the data points which are in one group are more similar to the other data points in the same group and more dissimilar to the other data points in the other group.

In simple words, we can say that we are grouping the data on the basis of the similarities and dissimilarities in the data.

Association in Machine Learning or Association Rule Learning

Association analysis of Association rule Mining is a technique in which we try to find interesting associations and relationships between items in the data. We are checking how one data item is dependent on another data item, or what is the connection here.

For example, this Association Rule mining is used a lot in Market Basket Analysis, which we can understand as if we are checking the association between the products like if someone buys milk and bread, they are more likely to buy eggs as well. So, this is something that you can predict from the data if it happens frequently, which can result in greater profits.

Anomaly Detection Machine Learning

If you try to google the term anomaly, you would find it similar to the term “outlier”, which is quite frequently used in Machine learning, especially when you are in the stage of understanding the data.

It basically involves detecting rare events, items, or some unusual observations, which significantly differ from the majority of data.

Anomalies are also known as outliers, which are simply the data points, which do not fit into the usual pattern or the trend that is observed in the majority of the data.

It can help us in many terms and has various applications, like Fraud Detection, Predictive Maintenance, etc.

3. Semi-Supervised Learning

As of our discussion now, you might be familiar with the terms Supervised learning and Unsupervised Learning. So, Semi–Supervised learning, as the name suggests, is quite a combination of Supervised learning, and unsupervised learning.

Here, some portion of the data is labeled, and the other portion is unlabeled. This technique is often used when we have a large dataset with unlabeled data, and it is quite not possible or too much expensive to label all the available data.

In such situations, we label some portion of data in the dataset, and the other one is unlabeled. So, we can understand this as the data learns from both the labeled and unlabeled data, to give predictions, or classify the items. It learns from the labeled data, and also, it has unlabeled data, which is used to discover new patterns and trends in the data.

4. Reinforcement learning

In supervised learning, we labeled data. In unsupervised learning, we had unlabeled data. In semi–supervised learning, we had labeled as well as unlabeled data, and now, in reinforcement learning, there is no concept like labeled data. The machine learns from experience.

This is like a machine has to explore the environment and perform some actions, and on the basis of that, it either gets rewards or is penalized. If it does the right thing, it is rewarded, and if it does something wrong, it is penalized. The goal here is to maximize the rewards, so the machine changes its policy accordingly and then takes action, so we can say that the machine learns from experience.

This is quite similar to how humans learn from their experiences, trying to do some different things, and observing the results, to get better.

Conclusion

In this article, we had a brief talk about some types of Machine learning, on the basis of the level of supervision required. There can be some more parameters of classification of machine learning, which you can explore in deep, but here, we could classify machine learning as Supervised learning, Unsupervised learning, Semi–Supervised learning, and Reinforcement learning.

We hope that you could understand at least the bigger picture of what is what, and now, you are ready to jump into something particular or explore the ocean of machine learning. We encourage you to learn and explore more about Python, Machine learning, Artificial Intelligence, etc.

FAQs related to Types of machine learning