Navigating Severe Imbalanced Class Issues in Predicting Students "At Risk" Using Cross-Industry Standard Process for Data Mining / Jose Silva-Lugo and Heather Maness.

Format:

Book

Author/Creator:

Silva-Lugo, Jose, author.

Maness, Heather, author.

Language:

English

Subjects (All):

At-risk youth--Research.

At-risk youth.

Data mining.

Physical Description:

1 online resource

Place of Publication:

London : SAGE Publications Ltd, 2025.

Summary:

The study provides a detailed methodological approach, cross-industry standard process for data mining, for predicting at-risk students with an imbalanced class. The objective was to identify the best machine learning model for predicting students at risk of failing the course during weeks 2-8 of the semester. We encountered issues in the dataset, including class imbalance, numerous missing values, and high variance. To address these issues, we oversampled failing students, removed missing values, and cleaned the data. Although the random forest model was the best among seven evaluated models, it overfitted and did not generalize well to the validation set. Despite various attempts to address overfitting with different techniques, further improvements can be made by addressing missing values and reducing noise. Readers will learn key techniques to mitigate issues with an imbalanced class, missing values, and overfitting. Furthermore, we demonstrate how to use sensitivity and specificity to determine the feasibility of intervening with a certain percentage of failing students. This methodology can be applied or adapted for similar research as it is flexible and allows for iteration among phases.

Notes:

Description based on publisher supplied metadata and other sources.

ISBN:

1-03-621730-2

9781036217303

OCLC:

1523169523