Types of Machine Learning and Associated Algorithms
So , from previous blog of “Technical Introduction and Process of Machine Learning” we have cleared out many things about ML and also understood thee process of how a ML model is created . Let us now jump to the next topic of the ML types and the what type of algorithms are used in those particular learning types .
TYPES OF MACHINE LEARNING
On the basis of the different types of problems , the type of data we collect and the functions we have to perform , a machine can be made to learn in different ways . These are —
1. Supervised Learning
Supervised learning is a type of process for a machine in which it is explicitly supervised about the data to learn from it . During the course of learning , the machine generates some equation and parameters so that it can do predictions in future .
Aim
In this type of learning , the aim is to train the machine in such a way that it can do predictions or forecast results in future .
Type Of Data
The type of data in supervised learning is always labelled . Its is similar to the way we were taught about various animals , fruits etc in our childhood . We were shown the photos of the animals along with their names i.e labelled with their names .
Type of problems
Supervised learning is basically used to solve two types of problems . These are —
- Classification Problems
Classification problems are those in which the machine is trained over a data having discrete output values.
Let us take an example of data that categorizes the fruit to be apple , orange or banana based on its shape , taste and color . Here , the output of the data has three categories which is a discrete number of values or we can say , a finite number of outputs . Hence , this type of problem is known as a classification problem .
Classification can be of two types —
- Binary Classification : Here , the data is classified into only two categories i.e “0” and “1” . For example : The data that classifies if a person is suffering from Cancer (1) or not (0) .
- Multi-class classification : Here the data is classified into more than two categories as explained in the above example of fruits .
- Regression Problems
Regression problems are those in which a machine is trained over the data having continuous output values .
For understanding the regression , let us take an example of salary to be given to the employees in the company . Here , based on different factors like working hours , festive bonus , medical and other deductions etc. , the salary of the employees is decided which is always a continuous value or we can say an infinite number of values .
2. Unsupervised Learning
Unsupervised learning is a type of process for a machine in which it is given the data as an input and then the machine has to find the hidden patterns in it to get trained .During the course of learning , the machine finds out the patterns or similarities in the input data values and it automatically groups them together .
Aim
In this type of learning , the aim is to discover all the hidden patterns in the data so that the machine can do the future predictions based on those patterns .
Type of Data
The type of data in the unsupervised learning is always unlabelled i.e the input data don’t have any kind of output category name given to it .
For example , if we are given a mixture of various types of grains in a bowl and not told what is the name of each grain present in the mixture . Now we have to separate each of the grains from others and put them in separate bowls .
Type of Problem
There are two types of problems that can be solved with Unsupervised Learning . These are —
- Clustering Problems
Clustering based are the problems in which the need is to divide whole data into a number of groups(clusters) or classes that are totally different from each other and the data in each class is mostly similar to its other classmates .
For understanding the clustering problem , let us take the above example of a mixture of grains give to us in the bowl . Here, we analyze the properties of each grain like colour and shape to separate them from others and put them in their respective bowls (clusters) .
- Association Problems
Association based are the problems in which the machine has to discover the relations of a certain object in data with other object and the chain them together .
For understanding the association problem , let us take an example of online shopping sites . You may have notices that whenever we buy a product from these sites , it shows a section of the window which says “ people who buy this product also buy” . This is an example of association as it shows that the product we have bought has an association to the other product .
3. Reinforcement Learning
Reinforcement learning is a type of process in which the machine(agent) gets trained by getting feedback from the environment when it performs any type of actions . Its is totally an action-feedback cycle to train a machine .
The basic idea of reinforcement learning can be taken from the real life example of training our pets like dogs or cats . When our dog (agent) stands on his two legs (action) and we (environment) want him to again repeat that and be good at it , we give him some reward like something delicious to eat . In this way , dog will learn that if he will do this action perfectly then he will get a reward and hence do that repeatedly .
Aim
In this type of learning , focus is on training the machine to perform a sequence of operations by adapting itself in the complex environment .
Type of Data
There is no data provided to the machine in this type of learning . Only an environment is created for the machine in which it gets trained.
Type of problems
The problems that can be solved by reinforcement learning are basically the different methods . These are —
- Positive Reinforcement
It is defined as a positive event or a condition that occur due to an action or behaviour of the machine . Here , a positive feedback is received from the environment when an action is done by the machine . This increases the strength and performance of the machine to perform that action .
For example : When you (agent) create a big project or achieve first rank in class (action) , you get rewards and congratulations from all the people around you (environment). This gives you a lot of confidence and encouragement .
However , a lot of positive reinforcement may lead to over-optimization of the action and may affect the results adversely . For example , you get over confident due to too much praises from your surroundings .
- Negative Reinforcement
It is defined as a negative event over a condition that occur due to an action or behaviour of the machine . Here , a negative feedback is received by the machine when it performs an action . This increases the strength and performance of the machine about not to do that action again .
For example : If you(agent) get failed in an exam(action) , you get negative feedback from your teachers and parents(environment) . This makes you strengthened to do studies thoroughly to avoid being failed the next time .
However, the drawback of this method is that it provides enough to meet up the minimum behavior . For example , after being failed , too much negativity and pressure over you can decrease your performance .
4. Semi-Supervised Learning
Semi-supervised Learning is a type of process that comes in between the Supervised and Unsupervised Learning . Here , the machine is trained over a small part of supervised data and a large part of unsupervised data.
In this type of learning , the machine is first trained in the form of supervised learning over a small data . After that , an unsupervised data is given to the same machine which is already trained over supervised data . The machine classifies the unsupervised data based its previous learning and generate results . All these results are added to the data and then again the machine is trained over whole data .
Type of Data
The type of data in semi-supervised learning is labelled and unlabelled both . The amount of labelled data is very less than the unlabelled data .
Associated Algorithms in Different Types of Machine Learning
1. Supervised Learning Algorithms
- Regression Algorithms : Linear Regression , Random Forest , AdaBoost
- Classification Algorithms : Logistic Regression , K-Nearest Neighbor(KNN) , Support Vector Machine(SVM) , Naive Bayes etc.
2. Unsupervised Learning Algorithms
- Clustering Algorithms : K means , Hierarchical Clustering etc.
- Association Algorithms : Association rules , Apriori etc.
3. Reinforcement Learning Algorithms
Q-learning , SARSA
Thank you .Hope this was an informative blog for you. Have a good day..!!