# Identifying At-Risk K-12 Students in Multimodal Online Environments: A Machine Learning Approach

**Authors:** Hang Li, Wenbiao Ding, Zitao Liu  
**arXiv:** [2003.09670](https://arxiv.org/abs/2003.09670)  
**Date:** 2020-03-21  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2003.09670v2/

## Key points

- K-12 online learning platforms have brought both opportunities and challenges to education
- Identifying at-risk students who may drop out of online courses is a challenge
- Previous studies focused on predicting dropout in higher education MOOCs, but few have developed machine learning approaches for K-12 online courses
- Hang Li, Wenbiao Ding, and Zitao Liu propose a machine learning framework that considers both online and offline factors to accurately identify at-risk K-12 students in multimodal online environments
- The approach addresses challenges posed by multiple modalities, length variability, time sensitivity, and data imbalance
- The authors conducted offline experiments on a real-world educational dataset and demonstrated that their method improves dropout prediction performance compared to state-of-the-art baselines
- They also tested their approach on a third-party K-12 online tutoring platform for two months and found that more than 70% of dropout students were detected by the system
- Preemptively identifying at-risk K-12 students is important to improve student retention rates and overall educational outcomes.

## Summary

The emergence of K-12 online learning platforms has opened up a new era of education, bringing with it both opportunities and challenges. One such challenge is identifying at-risk students who may drop out of online courses. While previous studies have focused on predicting dropout in higher education Massive Open Online Courses (MOOCs), few have developed machine learning approaches for K-12 online courses. In this paper, Hang Li, Wenbiao Ding, and Zitao Liu propose a machine learning framework that considers both online and offline factors to accurately identify at-risk K-12 students in multimodal online environments. The approach addresses the challenges posed by multiple modalities, length variability, time sensitivity, and data imbalance. The authors conducted offline experiments on a real-world educational dataset and demonstrated that their method improves dropout prediction performance compared to state-of-the-art baselines. They also tested their approach on a third-party K-12 online tutoring platform for two months and found that more than 70% of dropout students were detected by the system. This study highlights the importance of preemptively identifying at-risk K-12 students in order to improve student retention rates and overall educational outcomes.