TLDR: SentiDrop is a novel multi-modal machine learning model that predicts student dropout in distance learning. It uniquely combines socio-demographic data, behavioral data, and sentiment analysis of student comments using BERT and XGBoost. The model achieved 84% accuracy, demonstrating that integrating emotional insights significantly enhances dropout prediction and enables earlier, more targeted interventions to improve student retention.
Student dropout in distance learning is a significant global challenge, impacting individuals, educational institutions, and society. The flexibility offered by online education, while beneficial, also introduces complexities in retaining students. Traditional methods often focus on academic performance or absenteeism, but a new model, SentiDrop, proposes a more holistic approach by integrating diverse data sources to predict dropout risks early.
Developed by Meriem Zerkouk, Miloud Mihoubi, and Belkacem Chikhaoui, SentiDrop is a multi-modal machine learning model designed to identify students at risk of dropping out. What makes SentiDrop unique is its combination of socio-demographic data (like age, gender, economic status), behavioral data (such as login frequency and time spent on the platform), and crucially, sentiment analysis of student comments. This integration allows the model to capture not only objective indicators but also subjective elements related to students’ emotional experiences within the learning environment.
The core of SentiDrop’s innovation lies in its use of the Bidirectional Encoder Representations from Transformers (BERT) model for sentiment analysis. BERT is fine-tuned to understand the nuanced sentiments expressed in student comments, converting qualitative feedback into quantitative features. These sentiment scores are then merged with socio-demographic and behavioral data, which are analyzed using Extreme Gradient Boosting (XGBoost). The model also employs an ensemble approach, combining XGBoost, Random Forest, and Logistic Regression to enhance accuracy and robustness.
One of the key contributions of this research is highlighting the practical implications of sentiment analysis. By monitoring students’ emotional well-being in real-time, educational platforms can enable early and targeted interventions. Sentiment analysis can reveal underlying issues that traditional academic metrics might miss, such as disengagement or dissatisfaction, even if a student’s academic progression appears satisfactory. The study found that sentiments expressed early in the academic term have a disproportionately large impact on dropout likelihood, emphasizing the importance of early detection.
The SentiDrop model was tested on unseen data from a subsequent academic year and achieved an impressive accuracy of 84%, outperforming a baseline model that achieved 82%. It also demonstrated superior performance in other metrics like precision and F1-score. An ablation study further confirmed the significant impact of sentiment analysis, particularly for models like Naive Bayes and SVM, showing substantial improvements in recall and accuracy when sentiment data was included. While models like XGBoost showed less dependency on sentiment analysis, its inclusion still provided valuable insights.
Also Read:
- Unlocking Emotions Across Languages: A New Framework for Multi-Emotion Detection in Short Texts
- Improving Automated Essay Cohesion Scoring with Item Response Theory
This novel approach offers a fresh perspective on understanding and predicting student perseverance in online education. By providing more actionable insights, SentiDrop could be a vital tool for educators to develop personalized strategies, reduce dropout rates, and encourage student success. For more detailed information, you can refer to the full research paper: SentiDrop: A Multi-Modal Machine Learning model for Predicting Dropout in Distance Learning.
