Predictive Modeling of Learning Continuation in Preschool Education Using Temporal Patterns of Development Tests

Junpei Naito, Yukino Baba, Hisashi Kashima, Takenori Takaki, Takuya Funo

Introduction

Learning analytics has revolutionized higher education and online learning platforms by providing data-driven insights into student success. However, applying these techniques to preschool and primary education presents unique challenges. Predicting "dropout"—the point at which a child ceases supplemental education—is complicated by the subjectivity of parent-reported data, the scarcity of digital touchpoints in physical classrooms, and the fact that parents, rather than the students themselves, are the primary decision-makers. This research explores whether the temporal patterns within periodic developmental tests can serve as reliable predictors for learning continuation.

Data and Features

The study utilized a real-world dataset from the Shichida Educational Institute, consisting of developmental test scores for children in two age groups: 0-2 years (1,540 children) and 2-4 years (1,322 children). These tests evaluate four key domains: physical, perceptual, linguistic, and sociality development.

Developmental test scores over time across four domains: physical, perceptual, linguistic, and sociality.

Each test result is derived from parent-completed checklists, such as "Can the child jump over a 20cm height?" While these provide a rich longitudinal view of a child's progress, they are inherently noisy due to parental subjectivity. To mitigate this, the researchers extracted both "static" features (e.g., age at first test, total tests taken) and "temporal" features that capture the trajectory of development over time.

Methodology

To analyze the developmental trajectories, the researchers employed time-series clustering. This approach groups children with similar growth patterns, making the model more robust to individual measurement noise. A critical challenge in this analysis is that children take tests at different intervals and for varying durations.

To align these sequences for comparison using Euclidean distance, two interpolation techniques were tested:

  1. Extension method: Padding sequences with virtual data points to match lengths.
  2. Overlap method: Focusing exclusively on the time windows shared between sequences.
Interpolation methods for aligning time-series data: (a) Extension method and (b) Overlap method.

The researchers also utilized Dynamic Time Warping (DTW) to compare sequences of different lengths without explicit interpolation. These cluster assignments were then integrated into a Logistic Regression model to predict whether a child would continue their education past a specific age milestone.

Key Results and Insights

The inclusion of temporal patterns significantly improved the accuracy of dropout predictions compared to models using only static features:

  • Performance Gains: For the 0-2 year old group, the Area Under the Curve (AUC) increased from 0.7549 to 0.8001. For the 2-4 year old group, the AUC improved from 0.8093 to 0.8421.
  • Optimal Alignment: The "overlap" interpolation method combined with normalized Euclidean distance yielded the best results for the younger cohort, while DTW performed best for the older cohort.
  • Predictive Trajectories: The analysis revealed that the shape of development matters. Children who showed rapid score improvement, even if they started the program recently, were highly likely to continue. Conversely, regular testing combined with stagnant score growth was a strong indicator of impending dropout.
  • Behavioral Signals: Beyond the scores themselves, the frequency of testing was a vital signal; a decline in testing frequency often preceded a family's decision to stop the program.

Conclusion

This study demonstrates that temporal patterns in developmental test scores are powerful indicators of educational commitment in early childhood. By using clustering and sophisticated sequence alignment techniques, researchers can overcome the noise inherent in subjective, parent-reported data. Future work may involve applying more complex models like Recurrent Neural Networks (RNNs) and developing intervention strategies to support families identified as "at-risk" of discontinuing their educational journey.

Summary updated

Abstract

Learning analytics applies data analysis techniques to learning data in order to support students’ learning processes and to improve the quality of education. Despite the increasing attention to learning analytics for higher education, it has not been fully addressed in primary and preschool education. In this research, we apply learning analytics to preschool education to predict the continuation of learning of preschool children. Based on our hypothesis that temporal patterns in the assessment scores of development tests are effective features for prediction, we extract the temporal patterns using time-series clustering, and use them as the features of prediction models. The experimental results using a real preschool education dataset show that the use of the temporal patterns improves the predictive accuracy of future continuation of study.
BibTeX
@inproceedings{naito2018predictive,
  author = {Junpei Naito and Yukino Baba and Hisashi Kashima and Takenori Takaki and Takuya Funo},
  title = {{Predictive Modeling of Learning Continuation in Preschool Education Using Temporal Patterns of Development Tests}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.1609/aaai.v32i1.11393},
}
Figure