Enhancing a Probabilistic Auto-regressive Model with Gaussian Noise and Savitzky–Golay Filter for the Data Generation of Small-Scale Education Datasets

Sufficient samples are important to train accurate and robust machine learning models. In education datasets, common applications, including learning analytics, innovative learning, educational administration, and knowledge management, require models that can support small-scale training datasets. These applications limit the choices of algorithms, are prone to model overfitting and bias, and lack generalizability; thus, they possess room for research and enhancement. In this paper, a research work was presented on the data generation algorithm for the enhancement of small-scale education datasets. A probabilistic auto-regressive model was implemented and enhanced by merging it with Gaussian noise and Savitzky–Golay filter. The overall data distribution was improved, and more helpful training data is available. This approach also enabled the data generation of multiple variables simultaneously because incorporating the mutual dependence of variables enhances the characterization ability. Future research directions were also suggested.