Zachary Combs’ MS in Statistics Project Presentation

Date: Thursday November 30^th, 2023

Time: 8:30 am – 10:30 am

Location: ACAD-4018 & Zoom

Zoom Link: email mathstats-staff@ucdenver.edu for Zoom link

Committee Members: Erin Austin (committee chair & advisor), Joshua French, Emily Speakman

Title: Data Imputation with Artificial Intelligence

Abstract:  As we continue to progress through the information era, the need to collect data and be able to make accurate and informed decisions has become even more important in society. Organizations are using data to make business decisions. Politicians use data to appeal to their audience. The general citizen must make thousands of decisions based on the data they have to make it through a normal day. So, what if a portion of the data we had access to was missing?

With trillions of decisions being made globally at any given moment, there has never been more of a need for accurate and complete data. Thus, when data is missing, when we do not have access to the full story, we still need to make accurate decisions without introducing bias or artificially reducing the variance. However, with missing data, these decisions can be harder to make. Missing data can lead to inaccurate or misleading conclusions, making it necessary to know how to handle missing data when it appears. Data imputation is the broad term for how to handle missing data.

As machine learning algorithms become more advanced, the use of artificial intelligence to aid these algorithms has become popular. Artificial intelligence (AI) is an advanced version of machine learning, capable of learning from previous examples to make its own decisions. With the increasing power of technology, the use of AI has been recently introduced into the public view, increasing its popularity and accessibility, including in statistical fields. As the popularity of AI grows, so does its need to perform necessary tasks such as data imputation.

In this study, we will employ different data imputation methods and compare their performances to Akkio’s online AI tool. Classic imputation methods such as mean, regression and multiple imputation will be tested alongside this AI tool. Using the famous Iris Flower dataset, we will use different error measurement metrics to compare the performance of these tools. This paper will discuss if the current AI tools we have access to are capable of accurately predicting missing values in datasets with varying degrees of missing data and if not, whether they show promise for the future.