Limitations of Predictive Analysis

 

In my blog post discussing traditional statistics, I discussed the idea of inferential statistics and the limitations of using a sample to make estimations about the populations the sample is drawn from. In the world of Big Data this translates to the concept of predictive analytics. This is when Big Data leverages huge data sets, statistical algorithms, and machine learning to make predictions of future outcomes.

Predictive analytic models come in two types: classification and regression models. Classification models put data objects into categories and make predictions based on this whereas regressions models predict continuous data. A classification model may sort customers into categories and make predictions about how receptive they would be to marketing whereas a regression model would make predictions about how much money a customer will generate during their relationship with the company. (1)

The idea of predictive analytics is becoming an increasing part of Big Data, and the theory makes sense that with larger data sets or larger sample sizes then the predictions made on the analysis of these samples become more accurate. The issue is that no matter how large your data set is, it will always be a prediction and never a guarantee. Despite this, Business and organisations have become overconfident in the accuracy of large data dets when applying them to individuals.

Predictive Analytics, like all technologies, are a tool than can be used to gain insight into data but they are not a crystal ball which can be used to see into the future. This is the nature of all statistics, they are educated guesses when all is said and done, even if the educated guess is being made with exabytes of data backing them up.

 

One example of where predictive analytics failed is the Google Flu Trends. GFT was a service developed by google that aimed to use Big Data to analyse Google search query data to predict influenza activity in various regions of the world. The idea was that people would search certain terms when they were sick allowing the service to detect rising flu cases. This ultimately failed for a few reasons. One being because people search for these queries even when are not sick for various reasons and increased media coverage of the flu can lead to increased search activity creating a feedback loop. The algorithm also used historical correlations, but each flu season tends to be different and human behaviour changes. Therefore, changes in search behaviour, media attention, or even they language people use to describe their symptoms all evolves over time and affected the accuracy of the results. (2)

1. https://cloud.google.com/learn/what-is-predictive-analytics?hl=en

2. https://theconversation.com/googles-flu-fail-shows-the-problem-with-big-data-19363

Comments

Post a Comment

Popular posts from this blog

Value of Data

 In the modern world, data is a valuable resource. The value of data though can be harder to quantify compared to other more tangible resources such as equipment, real estate, or employees. Though these assets can be more easily digested in terms of cost vs. value, the impact data has had on the modern world in undeniable through cost reduction, revenue increase, and income generation. With data being infinitely produced at an ever-faster pace, the value of data is only increasing. Though data is being constantly produces in the modern, the type and quality of this data varies massively. Data alone doesn’t necessarily have the value that business and organisations are looking for and that is why Big Data technologies must be applied to extract value from the endless sea of data and gets that data to the right people at the right time. Using Big Data to extract value from data comes at a cost and how the end results balance out this cost can be understood through the 1-10-100...
Read more

What is Big Data?

 "Data is information that can be interpreted and used by computers. It is a collection of facts, such as numbers, words, measurements, observations or even just descriptions of things. In computing, data is typically stored electronically in the form of files or databases. Data can come from many sources including user input (typed words or images), sensors (temperature readings.) or algorithms (calculations)." (1) In today's modern world this data is being generated at such a scale and rate that it can no longer be processed by traditional data management tools. This is what is known as Big Data. It is estimated that 90% of the worlds data was created in the last two years alone with current trends estimating the amount of data created in 2025 will be 181 zettabytes. (2) These massive amounts of data being created can be categorised in three ways: Structured - This kind of data is highly organised, easily searchable, and is easily processed by Big Data technology in dat...
Read more

Traditional Statistics

  In my last post I discussed the value of data. In that post I discussed how value is extracted from data by processing raw data and packaging it but what does this mean? Data in its purest form is just points of information. Without context or analysis applied it doesn’t necessarily serve much use. One way in which human beings derive value from is data is through statistics. Traditional statistics is a tool for understanding and interpreting data. There are many ways in which this tool can be applied to data, with two main branches: descriptive and inferential . Both are ways to extract meaning from data but are used in different ways for different purposes. Descriptive Statistics Descriptive statistics are used to describe or summarise data in a meaningful way. They help us visualise what the data is showing and make it easier to recognise patterns that may emerge from the data. Descriptive statistics are applied to a sample and measures chosen properties of that sample t...
Read more