Machine learning and data analytics for multilayer data in policy planning

Vauhkonen, Mumtaz Bee

Permalink

https://hdl.handle.net/2142/97750 Copy

Description

Title

Machine learning and data analytics for multilayer data in policy planning

Author(s)

Vauhkonen, Mumtaz Bee

Issue Date

2017-04-21

Director of Research (if dissertation) or Advisor (if thesis)

Dhillon, Pradeep

Doctoral Committee Chair(s)

Dhillon, Pradeep

Committee Member(s)

• Cope, Bill
• Liang, Feng
• Koyejo, Sanmi

Department of Study

Educ Policy, Orgzn & Leadrshp

Discipline

Educational Policy Studies

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine learning
• Data mining
• Policy planning
• Multilayer data
• Education

Abstract

“Does bigger data lead to better decisions?” has been a frequent question for discussion among many decision makers—data scientists as well as organizational leaders and managers. Educational institutions, finance, and the retail industry have had big data for several decades that did not significantly alter decision making as it is doing today, as most of the decisions made were driven more by either small-scale studies or a desire to support a group’s belief or interests. However, the advent of computational mechanisms changed the notion of what big data analysis is (Evgeniou, Gaba, & Niessing, 2013). Linking these methods to diverse related data is proving to be a game changer for big data analysis; for example, connecting sales figures to customer behavior is leading to better business decisions. By extending this research into the education domain, data sources at multiple levels in education can be combined to connect the big picture by analyzing and connecting data at higher macro levels to lower levels. The main purpose of this study is to present a framework for analyzing macro data with multiple data types in education for effective policy planning by linking data analysis at multiple macro levels (district and school levels) and extend it demonstrate a proof of concept of connecting to micro levels. Achieving this task required the following steps: 1. Use an unsupervised approach in big data techniques to analyze data at school level and district levels. 2. Develop a supervised classification system to use clusters from step 1 as classes so that changes in school data are constantly reassessed and assigned to new classes. 3. Identify frequent patterns and association rules of various attributes at the school level. 4. Perform a regression analysis on results from the macro level to investigate deeper with additional variables to identify the differences and verify the frequent patterns and association rules. 5. Develop a simulated analysis of student-level performance data to identify similarity and dissimilarity patterns using collaborative models that gain insights into collective intelligence in a recommender system format, which teachers can use to find optimal measures to improve a student’s learning. These steps, put together into an analytical system, can handle large volumes of data and give insights for developing effective macro and micro policies. The results indicate that applying machine learning and data mining models enable extracting more insights at a macro policy planning level

Graduation Semester

2017-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/97750

Copyright and License Information

Copyright 2017 Mumtaz Vauhkonen

Owning Collections