Individual-Based Modeling and Data Analysis of Ecological Systems Using Machine Learning Techniques

Mashayekhi, Morteza. “Individual-Based Modeling and Data Analysis of Ecological Systems Using Machine Learning Techniques.” (2015).
URL1 URL2

Artificial life (Alife) studies the logic of living systems in an artificial environment in order to gain a deeper insight of the complex processes and governing rules in such systems. EcoSim, an Alife simulation for ecological modeling, is an individual-based predatorprey ecosystem simulation and a generic platform designed to investigate several broad ecological questions, as well as long-term evolutionary patterns and processes in biology and ecology. Speciation and extinction of species are two essential phenomena in evolutionary biology. Many factors are involved in the emergence and disappearance of species. Due to the complexity of the interactions between different factors, such as interaction of individuals with their environment, and the long time required for the observation, studying such phenomena is not easy in the real world. Using data sets obtained from EcoSim and machine learning techniques, we predicted speciation and extinction of species based on numerous factors. Experimental results showed that factors, such as demographics, genetics, and environment are important in the occurrence of these two events in EcoSim. We identified the best species-area relationship (SAR) models, using EcoSim, along with investigating how sampling approaches and sampling scales affect SARs. Further, we proposed a machine learning approach, based on extraction of rules that provide an interpretation of SAR coefficients, to find plausible relationships between the models’ coefficients and the spatial information that likely affect SARs. We found the power function family to be a reasonable choice for SAR. Furthermore, the simple power function was the best ranked model in nested sampling amongst models with two coefficients. For some of the SAR model coefficients, we obtained clear correlations with spatial information, thereby providing an interpretation of these coefficients. Rule extraction is a method to discover the rules explaining a predictive model of a specific phenomenon. A procedure for rule extraction from Random Forest (RF) is proposed. The proposed methods are evaluated on eighteen UCI machine learning repository and four microarray data sets. Our experimental results show that the proposed methods outperform one of the state-of-the art methods in terms of scalability and comprehensibility while preserving the same level of accuracy.

Related articles