A data mining technique for detecting emergent behaviors in agent-based simulations

The use of statistical techniques has become crucial for detecting and studying emergent behaviors in complex systems. Emergent behavior in complex systems often involves subtle or small changes in system dynamics or patterns. Statistical techniques are sensitive to small effects and can help detect and quantify them. Systems that show emergent behaviors usually involve multiple variables and data sources. Statistical techniques can integrate and analyze data from these diverse sources, allowing the identification of emergent patterns that may be scattered across different data sets. Therefore, this paper proposes the use of principal component analysis (PCA) for visualizing and detecting emergence. PCA is a mature and powerful unsupervised learning algorithm that allows the reduction and visualization of large datasets. The dataset set is obtained from simulations in NetLogo, and interaction metrics are used to represent the data. PCA was able to detect throughout the analysis emergence. In 90% of scenarios, PCA showed clear detection of emergent behavior in the system, the study also gives future directions for this work.