What is Machine Learning (ML)?
ML is a scientific study of algorithms and statistical models that computer systems use in order to perform a specific task effectively without using explicit instructions, relying on patterns and interfaces instead.
Types of Machine Learning:
The entire machine learning is divided into types based on the algorithms. Types are listed below:
This learning is responsible to direct an activity and check whether it is working in the desired manner or not, if it’s not working in desired manner make it correct to work properly. In this, machine starts learning under a set of rules/guidelines by feeding labelled data, passing input & output and train the machine to work on the data which we pass and machine starts working till it gives the output where it should match with the output which we passed while training the machine. This type solves two types of problems.
Classification is a problem where the output variable is a category such as colour, diseases, mails (Spam/not-Spam), etc.
Regression is a problem where the output variable is a real value/continuous quantity such as weights, heights, currency, rates, etc.
Example: Now, we are passing the below image as input and we try to find out how many count of each vegetable.
- Machine starts identifying and try to classify what are there in this image?
- Classification starts and it says these are vegetables.
- Now again it tries to classify again based on their shape and colour.
- How does it classify?
- Round and red colour – tomato
- Oval shape and mud colour – potato
Example: Every country in this world has their own currencies (inr, usd, euro, etc.) and the currency conversion between the countries is calculated daily and the output is a real value. We will pass the input daily and after some day’s, graph is plotted and check when and what is growth/downfall of the currency exchange rate and predict the further days conversion value.
Example: Calculate country’s GDP by passing many factors like bike sales, car sales, electronic sales, etc. and predict the GDP as an outcome.
Some of the most used algorithms under supervised learning are
- Support Vector Machines
- Linear Regression
- Logistic Regression
- Naïve Bayes
- Linear Discriminant analysis
- Decision trees
- K-nearest neighbour algorithm
- Similarity learning
- Neural Networks
- Random Forest
It is a self-organizing where there’s no training data or labelled data. The machine itself finds the patterns based on input’s data and derives output. This type solves two types of problems in general.
Association problem describes rules based on your input data.
Clustering problem solves groupings in data based on similarities.
Example: Most common thought process of human beings, people visit markets to buy groceries. Here it solves association problem like, if the customer buy salt, he/she tends to but peppers also. In this way we try to do associations between the customer thought processes. Many more examples like bread and jam, Bowl and spoon, etc.
Example: People buy goods from online, based on their purchase history we will separate them as clusters, like in any e-commerce web-site based on type of goods customer buy, we will form some clusters that some people belongs to fashion shopping and some belongs to electronic goods buyers, etc.
Some of the most used algorithms under unsupervised learning are
- K-Means algorithm for Clustering
- C-Means algorithm for Clustering
- Apriori algorithm for Association
- Association Rule mining for Association
- Principal component analysis
- Singular value decomposition
- Independent component analysis
It’s a self-learning where it allows machines to determine automatically about its behaviour by gathering the observations from the environment and make it interactive.
Example: Robots. Where they don’t know anything but after some actions it gathers some information and make use of that and finds the solutions where it maximizes the productive work and minimizes risks. Robots learns themselves by taking moves right/left/front/back and for every move it receives rewards or punishment. From this data, it learns by itself.
Some of the most used algorithms under Reinforcement learning are
- SARSA(State Action Reward and State-action) Algorithm
Machine Learning Use Cases:
Machine learning has a lot of unending list of use cases. Here are the few use cases.
- Business sectors
- Risk Evaluation
- Forecast Sales
- Risk analysis
- Predicting sales
- Predicting profit/loss
- Recommendation Systems (based on purchases in E-commerce web-sites)
- Anomaly detection (Credit-card Frauds)
- Self-driving cars
- Games (Ex: alphago)
As everyone knows, today’s world is running behind ML & AI, we (VirtueTech Inc) are also taking challenges to solve many issues which I discussed as examples above. Will come up with a new blog by taking a use-case and solve it by using any one of the above algorithms.