Training Versus Hiring in Security Operations Centers: Agent-Based and Reinforcement Learning Simulations of Training Duration Effects on CyberTeam Performance
Year: 2026
Type: preprint
Abstract: Security operations centers face critical resource allocation decisions: invest in training existing personnel or recruit additional staff? While team performance in cybersecurity depends on structure, skills, and cohesion, existing computational research has examined only static team composition without addressing how training duration affects performance trajectories. We address this gap by incorporating temporal dynamics into agent-based modeling of cyber team competitions, using reinforcement learning to simulate how teams develop strategies through sustained training.Our computational experiments ran thousands of episodes, which demonstrated three key findings. First, training duration affects performance nonlinearly through three phases: minimal improvement during early exploration (100-500~episodes, the win rate ranges from 0.00-0.15, rapid gains during skill acquisition (500-2,000~episodes, the win rate ranges from 0.53-0.57), and diminishing returns during refinement (2,000-5,000~episodes, the win rate ranges from 0.93-0.95). Second, optimal training investment occurs at 500-1,000~episodes, where marginal gains per additional episode are highest and returns decline substantially beyond 2,000~episodes. Third, trained teams achieve about 3-7$\times$ efficiency advantages in most scenarios (and higher in capped cases), with 5~trained attackers performing equivalently to roughly 15-35 untrained attackers.These findings provide quantitative evidence for cybersecurity workforce planning decisions. The nonlinear learning trajectory suggests organizations should design training programs to extend beyond exploration phases to capture high-yield improvements. The substantial efficiency advantages support training-focused workforce strategies, with a full 5,000-episode training program substituting for the recruitment of roughly 20-25 additional personnel.
Source: Research Square
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7166333383
