Developing AI-Powered Adaptive Testing Systems with Few-Shot Learning Education Techniques

Abstract

A few-shot learning is investigated as a method to develop AI-based adaptive testing systems for personalized learning in the field of education. The proposed model utilizes few shots learning to adapt the difficulty level of the tests according to limited information, and provides personalized tests to the students at real time to enhance learning results. The methodology includes data collection from various sources, including student behavior, assessments, and demographic data. It exploits preprocessing on the Open University Learning Analytics Dataset (OULAD) followed by the construction of a small number of examples in the dataset to allow the model to learn from a few-shot. The system utilizes AI-driven feature engineering to forecast student capabilities and choose the most suitable test items, developing an adaptive test flow. The results show high accuracy (99.91%), high macro average precision (99.92%) and high macro average F1-score (99.88%), indicating that the system can make accurate predictions even with small number of training examples. The results show that the system can always beat the traditional models, such as logistic regression (78%), random forests (84%) and gradient boosting (86%) through statistical analysis of the F1-scores in one-shot, five-shot, and ten-shot learning scenarios. The FSL-Adapt Test model had an F1-score of 0.901 for the one-shot tasks, 0.836 for the five-shot tasks, and 0.886 for the ten-shot tasks, and an impressive F1-score of 0.775 and 0.859 for the novel tasks (Novel F1 = 0.775 for five-shot and 0.859 for ten-shot tasks). The results validate the power of few-shot learning for scalable, efficient and adaptable testing toward better educational outcomes. The results highlight the promise of using AI and few-shot learning methods to transform adaptive testing systems.