Data Analytics Part 1 - from descriptive to prescriptive

[Music] in this the first of three videos we take a look at the world of data analytics arguably since humans started communicating we have had information to share and therefore to analyze and help make decisions today data analytics is closely linked with digital data as computer systems have become ubiquitous in business and in our wider lives the digital universe is not only a reflection of the real world it is becoming a real component within it it has enabled the design of new business models like Google uber and airbnb and is rapidly changing the way we interact with the world through technology like social media chat box and Virtual Assistants if you want to take part or perhaps even Drive this digital revolution you'll need an understanding of data analytics data analytics is about analyzing data to gain insights to help better decision making but we have a challenge our data is sitting inside databases which sit behind business applications for example s ap Oracle and Salesforce in fact it is likely to be in multiple places in multiple formats and running intensive analytics may slow down those applications and impact the business safely getting the data into your analytics tools will require some intermediate systems such as a data warehouse and data Mart's moving the data between these systems will require what's known as a data pipeline collectively this is called a data infrastructure and it is a significant factor in every analytics project our end goal is to make better decisions these decisions will affect our future whereas the data available to us is based on the past this is a problem which is driving the evolution of data analytics it leads to the concept of analytics maturity levels in its simplest form there are three levels within the analytics maturity model an organization's ability to carry out analytics must start at the bottom with descriptive analytics the higher levels cannot be attained without this in place and indeed progression must be through each level in turn predictive and prescriptive analytics are the realms of advanced analytic technologies such as machine learning and artificial intelligence the practice of using data to predict the future is called data science these practitioners are called data scientists they're not called data Wizards for a reason descriptive analytics is about deriving insights from historical data it's a bit like a rearview mirror it's highly useful but you would not want to drive your car using it data can include point-in-time views right up to the present for example sales figures product mix and margin contribution it can also include trends which are a series of data points plotted against time nowadays there is a myriad of software tools which can combine and present data in colorful and interesting ways data visualization has become big business but ultimately this is simply presenting views of the business based on historical data now you may say surely we can use historical trends to predict the future maybe maybe not imagine house prices have been rising on average at 6% per year for the last 40 years and indeed in the last couple of years it's an even steeper upward trend would you feel confident investing money in housing this was the case in 2007 however the world discovered that in fact past performance does not predict future performance housing markets can crash as can fish stocks crop harvests and manufacturing yields in 2007 the housing market crashed big time and in its wake it caused a banking crisis which is still affecting the world economy a decade later this is something that an extrapolation of a trendline would have missed completely predictive analytics is forward thinking it is about predicting the future then using this information to make better decisions so how is this different from simply extrapolating a trendline the answer is modeling if you could build a digital model which accurately reflects a real-world system then you could use that model to simulate the behavior of that system into the future with today's computing capabilities even complex models are used routinely to test systems digitally before they are created in the real world for example designing buildings towns and cities planning the travel paths of spacecraft across our solar system and using flight simulators to allow pilots to safely learn how to handle extreme situations model building can be achieved through mathematical formulas if you know the parameters that define the system for example to model a ball falling through the air you simply need to know the gravitational constant the weight of the object and its resistance to travel in air it's just maths easy right having an oversimplified model can cause predictions to fail what happens when the ball hits the ground and what if instead of one ball we have many we also need to consider random factors like gusts of wind or objects lying on the floor George box famously captured this issue with modeling when he said essentially all models are wrong but some are useful the question is around how wrong the model is and whether it is good enough provide valuable and reliable insights there are situations when we don't know the parameters which define the system and so we cannot build a model based on equations instead we could use machine learning to build a profile of its behavior the concept is that we start with data about a real-world system the more the better and we use it to create a model that approximates it the model can predict an output then we introduce machine learning routines to iterate the model and improve it this is achieved by inputting historical data to the model then comparing its outputs to known outputs from the real world system based on the differences the routines modify the model and retest this approach is used iteratively over time constantly testing the models behavior against the real world system and making improvements and so the accuracy of the model is close enough to reality once the model has sufficiently learned how the real world system behaves it can be used to predict behaviors prescriptive analytics is also forward-thinking and builds upon predictive capabilities the difference is that with prescriptive analytics we use machines to start making recommendations by providing some criteria for what is a good or bad outcome prescriptive algorithms could determine how to achieve the best outcome in some cases we can enable the machine to execute on the recommendations

Channel: 2DeCipher

Views: 12,345

Keywords: data analytics, data pipeline, data infrastructure, descriptive analytics, predictive analytics, prescriptive analytics, data science, data warehouse, data mart, machine learning, artificial intelligence

Id: vax2bgG8hu8

Length: 8min 9sec (489 seconds)

Published: Sun Feb 03 2019