GA-GWNN: Detecting anomalies of online learners by granular computing and graph wavelet convolutional neural network

As online learning is becoming popular, detecting anomalous learners is crucial in improving the quality of teaching and learning. Such anomalies are hidden at different granularity levels of data. However, most relevant existing approaches fail to use the data at different granularity levels. To address this problem, we propose a novel framework called the Granularity Adaptive-Graph Wavelet Neural Network (GA-GWNN) to detect anomalies of online learners. The basic idea of GAGWNN is to fuse graph convolutional neural networks and granular computing techniques to mine different types of online data at multiple granularity levels. Specifically, the important features are first selected using a rough set and an attribute reduction algorithm. Different types of data are then granulated to obtain their corresponding information grains of the selected features. A weighted undirected graph is finally constructed where the nodes are features and the weights of the edges reflect the degree of these feature relationships. With this graph, we aggregate the nodes using the Louvain algorithm, build the hierarchy of the aggregate graph as the first part of GA-GWNN, and restore the aggregate graph to the original graph in the second part of GA-GWNN. By using both local and global information hidden in a data set, GA-GWNN can derive knowledge about both learners and the groups to which they belong at different levels of granularity. The experiment results demonstrate that GA-GWNN can effectively detect anomalies of online learners, such as losing concentration and a decline in learning performance. A comparison of the results of the experiments on five real-world data sets shows that, on average, GA-GWNN achieves a 1.5% improvement over several state-of-the-art methods in terms of precision, recall, and F-measure.