FORMATION OF DECISION-MAKING STRATEGIES IN BUSINESS ECOSYSTEMS THROUGH EVOLUTIONARY SEARCH BASED ON GENETIC ALGORITHMS

Abstract

In the context of growing complexity and dynamism of modern markets, effective strategic decision-making within business ecosystems requires computational models capable of adapting to uncertain, resource-constrained, and multi-agent environments. This paper presents an evolutionary approach to decision strategy formation using genetic algorithms, designed to model and optimize behavior in simulated business ecosystems. The proposed model treats the ecosystem as a population of heterogeneous agents with interdependent strategies, competing objectives, and dynamic interactions. Each agent operates under individual constraints related to resources and risk thresholds, while the system optimizes a global fitness function composed of profitability, stability, and inter-agent balance.

A specialized virtual environment was developed to simulate multi-agent dynamics and visualize strategy evolution. Agents are represented as network nodes whose behavior is encoded into chromosomes. The evolutionary engine utilizes tournament selection, single-point crossover, and Gaussian mutation. Fitness evaluation accounts for both local and systemic goals, and a feasibility check penalizes unfit solutions. Simulation results over 100 generations showed fast convergence, with significant improvements in average and best fitness, and a marked decrease in strategy variance. The genetic algorithm was benchmarked against greedy heuristics and random search, demonstrating superior performance in terms of solution stability, adaptability, and overall effectiveness.

The study highlights the advantages of genetic algorithms in modeling emergent behaviors and adaptive strategy formation in business ecosystems. However, limitations include the reliance on synthetic data and fixed algorithm parameters. Future work will explore hybrid evolutionary-learning models and real-world validation using business case data to enhance realism, scalability, and domain applicability.