A Weighted Feature Extraction Technique Based on Temporal Accumulation of Learner Behavior Features for Early Prediction of Dropouts
| Autor: | Andy W. H. Khong, Sivanagaraja Tatinati, Kai Liu |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Exploit
Computer science business.industry Massive open online course 05 social sciences Feature extraction Preemption 050301 education 020206 networking & telecommunications 02 engineering and technology medicine.disease Machine learning computer.software_genre Task (project management) Correlation 0202 electrical engineering electronic engineering information engineering medicine Attrition Artificial intelligence business 0503 education computer Dropout (neural networks) |
| Zdroj: | TALE |
| DOI: | 10.1109/tale48869.2020.9368317 |
| Popis: | Dropout prediction is an important task due to the high attrition rate commonly found on the massive open online course (MOOC) platforms. Performing accurate prediction leading to timely intervention is therefore important to reduce attrition. To address this challenge, we propose a feature generation approach to allow machine learning models exploit current learner behavior data to predict dropouts during their learning journey. The proposed feature generation approach analyzes the behaviors across time for each learner and determines appropriate weightings of the behavior for each time slice based on both recency and correlation, allowing existing machine learning models to extract patterns from the varying behaviors across learners. We evaluate the feasibility via various machine learning algorithms that employ the proposed generated features. Results show that the proposed techniques achieve higher accuracy in the early weeks compared to existing feature generation approaches. In addition, we demonstrate the feasibility of implementing the techniques in real-time machine learning pipelines for actual use. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|