Copyright © 2018 DataScience.US All Rights Reserved.
Predictive Analytics vs Machine learning
Big data holds a lot of value that businesses can analyse and interpret to make cost savings, improve their business and drive operations forward.
However, in order to access the wealth of information available the data needs to be processed in order to offer insights and different frameworks can have an impact on effectiveness.
Predictive analytics evolved from basic descriptive analytics. While descriptive analytics summarises large amounts of data into digestible chunks, such as showing sales for a set period or the amount spent on a particular outgoing, predictive analytics attempts to use this current and historical information to predict the future.
This can be as straightforward as receiving sales projections to using the patterns identified to anticipate how specific conditions could have an impact, for instance, how sales would be affected if advertising was increased or decreased. One of the most well-known applications for predictive analytics is credit scoring, where financial companies use a range of information to assess the likelihood of future credit payments being made.
Predictive analytics gives businesses the tools they need to plan for the future based on the probability of different scenarios happening. Through identifying potential opportunities and risks businesses can optimise their operations, making the predictive analytics process a lucrative process for businesses that effectively use the insights gained.
There are drawbacks to predictive analytics though, one is the amount of data it requires as the more historical data there is the more accurate the suggested outcomes will be and the need for this information to be kept up to date. While this wouldn’t be an issue for some business, smaller firms may not see as useful information as a result. Predictive analytics also focuses on past patterns and today’s digital and connected world means everything, including customer behaviour, is changing faster.
Machine learning is a method used to devise complex models and algorithms that then led themselves to prediction. This tool allows computers to learn without being directly programmed, enabling computer programs to evolve and adapt as new data is added. The computer will learn from previous calculations to produce reliable results.
Machine learning has plenty of applications from learning what a customer wants and making recommendations based on its conclusions to detecting what customers are saying about a business on social media platforms. The result of machine learning is high-end predications that can guide decisions in real time without the need to rely on human intervention.
Similar to predictive analytics, machine learning also has large data requirements. However, as machine learning doesn’t require human intervention it can prove to be a cost effective tool that offers scalable predictive analytics by using fully automatic methods, simplifying some typical data tasks.
Machine learning algorithms can broadly be segmented into supervised and unsupervised versions.
Supervised machine learning algorithms can be used for predictive analytics.
In supervised machine learning, we train the algorithm with labelled training set, e.g. given the weather conditions of 100 days and wanting to model which days would rain, we then choose an algorithm and feed it data about what the conditions were like on days where it rained, and on days where it did not rain. This trained model can then be applied on new data to predict if it will rain.
But not all predictive analytics needs to be done using machine learning approaches.
So the other side of it is predictive analytics techniques fall largely into two camps – besides machine learning techniques there are also a large family of statistical modelling techniques such as time series forecasting or various forms of regression.
The main difference between the two techniques are the assumptions of the data generation process.
In the statistical modelling approach, the assumption is that data is generated from a consistent but random data generation process. Hence the term ‘modelling the data’, and assuming various distributions etc. In machine learning, the is no such assumption. The data generating process is a black box.