Basic Machine Learning Concepts

Machine Learning technologies are making inroads into the commercial sphere after finding traction in academia. They may now be used by anyone to put their data to work and get competitive advantages that were previously only available to huge corporations and institutions.
Some ideas and basic concepts of Machine Learning to help in its understanding for those who have just landed in this exciting world.



1. Supervised and unsupervised machine learning-
There are two types of machine learning: supervised learning and unsupervised learning. Although the first appears to refer to human intervention in prediction while the second does not, the two notions are more closely tied to what we intend to do with the data.

One of the most common applications of supervised learning is to generate future predictions based on behaviours or features observed in previously recorded data (historical data). By linking all fields to a single field called "the target field," supervised learning makes it feasible to search for patterns in past data. Users, for example, categorise e-mails as "spam" or "genuine."

The study of which features or patterns the emails that have already been designated with both tags have begins the prediction process. For example, spam email can be identified as email that originates from specific IP addresses, has a specific text/images relationship, contains specific terms, has no one in the "To:" field, and (so many "and also" more)... This is only one of the possible patterns. New messages that have never been designated as spam or genuine are compared to patterns and classed (defined) as "spam" or "legitimate" depending on their features once all patterns have been determined (this phase is termed "learning").
Unsupervised learning, on the other hand, uses historical data with no target field. The goal is to look at the data and see whether there is any structure or organisation to it. It's frequently used, for example, to group clients who share features or habits with highly segmented marketing campaigns.

2. Classification and regression

These are supervised machine learning ideas. A classification system predicts a category, whereas a regression system predicts a numerical value.
The previously described spam is an example of categorisation. Emails are classified as either "spam" or "genuine." Another classic machine learning example of categorization is the prediction of churn, such as in a telephone firm. In this scenario, the goal is to detect consumer behavioural patterns that will be utilised to predict whether they would go to the competitor. Customers are categorised as "churn" or "no churn" in this situation.
The regression, on the other hand, forecasts a quantity, such as the price of an item or the number of reservations made in a given period.
3. Data mining
 
Data mining and machine learning techniques are frequently employed in a variety of ways. These are closely related ideas. The fundamental difference, in our opinion, is the purpose of each discipline. Machine learning is used to recreate known patterns and generate predictions based on patterns, whereas data mining finds previously unknown patterns.
 
In a nutshell, data mining serves as an exploratory tool, whereas machine learning focuses on prediction.
 
4.Learning, training
The heart of machine learning is the technique of detecting patterns in a data collection. With new data entered into the system, predictions can be generated once trends have been detected.
 
For example, historical data from online book purchases can be used to analyse customer behaviour in their purchasing processes (titles visited, categories visited, purchase history...), group them into behavioural patterns, and recommend purchases to new customers who follow known or learned patterns.
5. Dataset
It is the predictive system's raw material. This is the historical data that was used to train the pattern detection system. The dataset is made up of instances of factors, traits, and properties.
 
6. Instance, sample, record
 
Each piece of data available for analysis is called an instance. Each instance would correspond to a subscriber if you wanted to anticipate the behaviour of telecommunication service users. Each instance is made up of characteristics that describe it, such as the customer's age at the company, the amount of money spent on calls each day, and so on. The rows in a spreadsheet represent the instances, whereas the columns represent the qualities.
7. Feature, attribute, property, field

These are the attributes that each of the dataset's instances has. Depending on the author and circumstance, names are used interchangeably. In the case of a customer portfolio, we'd be talking about the number of purchases made by each client, their age, whether they're on social media, if they've signed up for the newsletter, what products they bought, and so on. Columns are the rows in a spreadsheet.

8. Objective

It is the attribute or component that we wish to predict, as well as the prediction's goal, such as the likelihood of a patient's readmission following surgery.

9.Feature Engineering

This is the step before creating the prediction model in which the data fields are analysed, cleaned, and structured. One of the most significant and costly prediction processes is this one. The goal is to delete fields that do not aid in prediction and appropriately organise them so that the model does not acquire information that is not relevant and potentially result in low-quality or low-confidence predictions.
In a nutshell, it is the process of removing noise from a signal.

10.Model

A model is constructed to generate predictions after the system has been trained (that is, after patterns in the data have been detected). A model can be compared to a filter into which new data is fed and the output is the classification of that data based on the patterns discovered during training. If a model is trained with past data to detect the danger of credit card cancellation, for example, the model will classify new consumers based on their behaviour to anticipate the cancellation. There are two types of machine learning: supervised learning and unsupervised learning. Although it may appear that the first refers to human intervention in prediction while the second does not, the two notions are more closely tied to what we intend to do with the data.