Automatic Assessment Item Bank Calibration for Learning Gap Identification

The benefit of assessments (formative or summative) is fully realized when they result in accurate identification of learning gaps. AMrita Personalized Learning and Evaluation (AMPLE) platform adopts a unified approach to pinpoint learning gaps by integrating student performance data from paper-and-pen assessments and computerized adaptive tests. In order to aid the estimation of learning gaps, we have also developed a calibration technique for automatic assignment of difficulty levels to assessment items. This calibration technique uses statistical features derived from student performance data and a Gaussian Mixture Model (GMM) to effectively identify difficulty levels of assessment items. We also show our verification of this model using a diverse data set of student assessments spread over six subjects and 6000 students. Our model achieved about 91% accuracy by comparing the model-generated output with teacher-supplied difficulty levels. By auto-calibrating the assessment items, we pave the way for accurate learning gap analysis, obviating significant efforts from the teacher.