Heuristic Algorithm for event prediction in Embedded Systems

This paper presents the results of a comparative analysis of methods without the use of neural networks (Weighted Random Walk, Specific Pattern Search, and 3rd order Markov chain), graph convolutional neural network (GCN), and the proposed heuristic algorithm for the task of predicting the next node in weighted oriented matrices. A heuristic algorithm was also proposed that uses a combination of edge weights and node characteristics. For each potential next node, the following information is collected: edge weight, node outdegree (number of outgoing edges), node indegree (number of incoming edges). Based on this information, a prediction is made. The considered weighted matrix is the output information of the embedded system, representing an updated model of the processes of the embedded system itself. The analysis of the obtained data is assumed outside the system itself. The experiments use synthetic data. In the study, matrix sizes from 5x5 to 300x300 with 20 iterations for each proposed algorithm are used as initial data. As a result of the comparison, the following metrics are obtained: prediction speed, accuracy, required memory, and training time. Recommendations for the application of the methods discussed are given, and the effectiveness of the proposed heuristic algorithm is also shown. The work serves as a practical guide to choosing an approach in the tasks of analyzing the structure of a graph built based on a weighted adjacency matrix.