Interpretable Neural Network Models in School Teacher Success Prediction

In this paper, we demonstrated the application of artificial neural networks in the field of educational data mining. The study is based on the data obtained from the large educational database, containing detailed information (more than 2 billion information units) describing educational processes in more than a thousand schools. Initial preprocessing of raw data was performed by using Big Data methods based on Dask framework. We developed predictive models, based on artificial neural networks. The models are represented by various feed-forward architectures built on the basis of PyTorch framework. In this work, we solved a specific problem, aimed to build a model for predicting the success of a school teacher, based on information about the educational process. The most optimal neural network-based architectures for solving this problem were obtained. To understand the features of trained models, interpretation methods based in Integrated Gradients and SHapley Additive exPlanations (SHAP) were used. The developed system allows to effectively predict and to determine the strengths and weaknesses of the educational process in schools.