# Data Analytics Popular Algorithms Explained

December 23, 2017 The world of Data analytics is constantly evolving, almost all manual repetitive tasks are being automated, and some complex ones too. If you are in the profession of big data, a data scientist, or from the field of machine learning, understanding the functions of these algorithms would be of great advantage.

A continuation of the earlier blog, mentioned below are a few popular algorithms commonly used by the data scientists and the machine learning enthusiast. The headings might differ slightly in terms of the nomenclature of the algorithms, but here we have tried to capture the essence of the model and technique.

## Linear Regression

Imagine you have many logs to stack together from the lightest to the heaviest, however you cannot weigh each log, you need to do this on appearances, the height and the girth of the log, only using the parameters of the visual analysis, you should arrange them. This, in other words, is Linear Regression, where a relationship is established between independent and dependent variable by arranging them to a line. Another example would be modelling the BMI of individuals using weight. You should use linear regression if there is a possible relationship or some sort of association between variables, if not then applying this algorithm will not provide a useful model.

## Logistic Regression

Just like any other regression, logistic regression is a technique used to find an association between a definite set of input variables and an output variable. But in this case, the output variable would be a binary outcome, i.e. 0/1, Yes/No, for e.g., if you want to assess, will there be traffic at Colaba, the output will be a specific Yes or No.  The probability of traffic jam in Colaba will be dependent on, time, day, week, season etc…, through this technique you can find the best fitting model that will help you understand the relationship between independent attributes and traffic jam, incidence rates, and the likelihood of an actual jam.

## Clustering

This is a sought of unsupervised learning algorithm where a data set is clustered into unique groups. So if you have a database of 100 customers, you can internally group them into different clusters or segments based on variables. If it’s a customer database that you are working on, then you can cluster them basis, gender, demographics, purchasing behaviour etc…, This is unsupervised as the outcome is unknown to the analyst. The algorithm is deciding the outcome, and an analyst is not training the algorithm on any past input. There is no right or wrong solution in this technique, business usability decides the best solution. There are two types of clustering techniques, Hierarchical, and Partitional. Clustering is also referred to by some as Unsupervised Classification ## Decision Trees

As the name suggests, decision trees is a visual representation of a tree-shaped visual, which one can use to reach to a desired or a particular decision, by simply laying down all possible routes and their consequence or occurrences. Like a flow chart for every action, one can interpret what would the reaction be for selecting the said option.

## K-Nearest Neighbors

The data science community essentially uses this algorithm to solve classification problems, although it can be used to solve regression problems as well. This algorithm is very simple, it stores all available cases, and then classifies any new cases by taking a vote from its K-Neighbours. The new case is then assigned to the class with the most common attributes. An analogy to understand this would be, the background checks performed on individuals to gather relevant information.

## PCA

The main objective of the Principal Component Analysis is to analyse the data to identify patterns and find patterns, to basically reduce the dimensions of the dataset with minimal loss of information. The aim is to detect the correlation between variables. This linear transformation technique is common and used in numerous applications =, like in stock market predictions.

## Random Forest

In the random forest, there is a collection of decision trees, hence the term ‘Forest’, here to classify a new object based on attributes, each tree gives a classification and that tree votes for that class. And overall the forest chooses the classification having the most votes, so in the true sense every tree votes for a classification.

## Time Series / Sequencing

Time series is an algorithm which provides regression algorithms that are further optimized for forecasting of continuous values, like for example, the product sales report, over a period of time. This model can predict trends based on the original dataset which was used to create the model. To add new data to the model, you need to make a prediction and automatically integrate the new data in the trend analysis.

## Text Mining

The objective of the text mining algorithm is to derive high-quality information from the text. It is a broad term which covers a variety of techniques to extract information from unstructured data. There are many text mining algorithms available to choose from based on the requirements. For example, first is the Named Entity Recognition, in which you have the Rule-Based Approach, and the Statistical Learning Approach. Second is the Relation Extraction, which further has, Feature Based Classification, Kernel Method.

## ANOVA

One-Way-Analysis of Variance is used to analyse if the mean of more than two groups of the dataset is significantly different from each other. For example, if a marketing campaign is rolled out on 5 different groups, where an equal number of customers are present within the same group, it is important for the campaign manager to know how differently the customer sets are responding so that they can make amends and optimize the intervention by creating the right campaign. The Analysis Of Variance works by analysing the variance between the group to variance within the group.

Optimise your knowledge by understanding these algorithms intensely if you wish to flourish in the field of data science.