3.8 Proceedings Paper

Balancing Fined-Tuned Machine Learning Models Between Continuous and Discrete Variables - A Comprehensive Analysis Using Educational Data

Journal

ARTIFICIAL INTELLIGENCE IN EDUCATION, PT I
Volume 13355, Issue -, Pages 256-268

Publisher

SPRINGER INTERNATIONAL PUBLISHING AG
DOI: 10.1007/978-3-031-11644-5_21

Keywords

Neural networks; Tree-based algorithms; Educational data mining; Feature engineering; MOOCs

Ask authors/readers for more resources

This study explores methods to handle challenging continuous variables datasets and predicts student dropout in MOOCs. A fair comparison methodology is proposed, and the results show that discrete models outperform time-series models in predictive ability and have faster training and testing speed.
Along with the exponential increase of students enrolling in MOOCs [26] arises the problem of a high student dropout rate. Researchers worldwide are interested in predicting whether students will drop out of MOOCs to prevent it. This study explores and improves ways of handling notoriously challenging continuous variables datasets, to predict dropout. Importantly, we propose a fair comparison methodology: unlike prior studies and, for the first time, when comparing various models, we use algorithms with the dataset they are intended for, thus `like for like.' We use a time-series dataset with algorithms suited for time-series, and a converted discrete-variables dataset, through feature engineering, with algorithms known to handle discrete variables well. Moreover, in terms of predictive ability, we examine the importance of finding the optimal hyperparameters for our algorithms, in combination with the most effective pre-processing techniques for the data. We show that these much lighter discrete models outperform the time-series models, enabling faster training and testing. This result also holds over fine-tuning of pre-processing and hyperparameter optimisation.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

3.8
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available