Machine Learning is becoming most trending field in 21st century. So, let us start with discussing what is machine learning.
Machine Learning can be considered as concept of feeding data to the machine(computer) and letting it learn of its own without being explicitly programmed. This sound more like human brain trying to train to accomplish a certain new task. This definition will be much clearer with a scenario- So, let us consider a human child, and that child is told to differentiate between cherry, oranges and apples.
He/she won’t even have an idea where to begin with. But as that child grows up, he/she acquires information in the brain to distinguish between fruits. The only reason why he/she was able to make this distinction is because he/she observes the surroundings, gathers data and learns from the past experiences. This is what exactly happens in machine learning as well, feeding data to a machine. The machine interprets the data, analyses it from past observation(experiences) and finds out patterns and key features to solve problem.
Now that we have clear insight about the definition of machine learning, hence let us move towards the types of machine learning approaches out there
- Supervised Learning
- Un-Supervised Learning
In this approach a learning function maps input variable “X” to an output variable” Y” based on previously observed input-output pair. Labelled data set is used in this approach i.e. each input variable has corresponding output variable or “Label” to it in order to set a boundary on which the algorithm should work. It can also be considered as “spoon-fed” version of machine learning as we select kind of output to feed the algorithm in order to train it and also the result we desire i.e. yes/no or true/false.
So, the purpose of supervised learning algorithms is to analyse the training data from the data set and produce a mapping function which will be further used to map new inputs/examples/observations. It can be considered as if a teacher is supervising students to achieve good marks in the coursework so they can pass. Here, student is the input which is supervised and processed by teacher (Mapping function) in order to obtain a specific result.
Supervised Learning algorithms are further divided into two approach:
These are supervised machine learning algorithms which categorizes entire data set into different classes based on various past data observations. Ex: classify emails as span and not spam.
These are supervised machine learning algorithms which facilitate us to find continuous variable based on multiple predictor values.
Following are the most widely used supervised algorithms:
- Decision Tree
- Random Forest
- Support Vector Machine (SVM)
- Linear Regression
- Multiple Linear Regression
- Naïve Bayes
- K-Nearest Neighbors
Few Common Use Cases for Supervised Machine Learning
- Sentiment analysis of social media post to understand user feedback
- Sales performance by tracking all data related to sales, email, calls and CRM interaction
- Human Resource allocation
- Time series forecasting to predict markets and future prices of products
- Security purposes in terms of spam filtering, fraud detection etc.
Unsupervised learning is category of machine learning approach which deals with finding a pattern in the data under observation. In this approach input variables “X” are specified without actually providing corresponding mapped output variables “Y”. So, it can be inferred from above line that the data set used for unsupervised machine learning algorithms unlabeled. It can be considered as if vendor arranges new items with respect to brand, type and quality in a shopping mall.
Un-supervised learning algorithm is used to find key features, patterns and to explore the structure in the data. Resulted patterns deduced on the basis of poor approximation as compared to supervised learning. However, it can also be used when we don’t have clear idea about the desirable output to the problem.
Un-supervised Machine learning is further divided mainly into two approaches:
It is a type of unsupervised learning approach in which entire data set is divided into various groups or clusters. In simple terms, crux of this approach is to segregate input data with similar traits into clusters
It is a type of machine learning technique which deals with reducing input variables in order to make predictive modelling easier. Noise in the data can make can make prediction process a lot more difficult, therefore dimensionality reduction can also be used to remove noise as well.
Following are the most widely used un-supervised algorithms:
- Principal Component Analysis
- Agglomerative hierarchical clustering
- Divisive hierarchical clustering
- DBSCAN clustering
- K-means clustering
- OPTICS algorithm
Few Common Use Cases for Un-supervised Machine Learning
- Recommendations on social media, streaming services etc.
- In the field of biology and medical sciences for genetic clustering and distinguishing between various cells and tissue types
- In the market to predict likelihood of item A being purchased by the customer after item B is already acquired
Now, you might have got an idea about Machine learning and its types.
That’s all for this post, thank you so much for reading. Follow up for related upcoming articles.
You can also check out our post on: Unsupervised learning with Python | Supervised and Unsupervised Learning