Optimizing Fantasy Sports Team Selection with Deep Reinforcement Learning

TL;DR

This work tackles the problem of optimizing fantasy cricket team selection by casting it as a sequential decision-making task solved with deep reinforcement learning. The authors formulate an OpenAI Gym-compatible RL environment where actions swap players between a current eleven and a reserve group, guided by a reward structure that incentivizes efficient convergence toward high-scoring configurations. They compare Deep Q-Network (DQN) and Proximal Policy Optimization (PPO) architectures, finding that PPO consistently yields higher-quality teams than baselines and DQN across cross-validation folds. The results demonstrate that RL-based team construction can shift predicted performance distributions toward higher percentiles, offering a data-driven tool with practical implications for user engagement and competitive play in fantasy sports. The approach is extensible to other sports domains and can be enhanced with real-time data integration and hybrid RL methods.

Abstract

Fantasy sports, particularly fantasy cricket, have garnered immense popularity in India in recent years, offering enthusiasts the opportunity to engage in strategic team-building and compete based on the real-world performance of professional athletes. In this paper, we address the challenge of optimizing fantasy cricket team selection using reinforcement learning (RL) techniques. By framing the team creation process as a sequential decision-making problem, we aim to develop a model that can adaptively select players to maximize the team's potential performance. Our approach leverages historical player data to train RL algorithms, which then predict future performance and optimize team composition. This not only represents a huge business opportunity by enabling more accurate predictions of high-performing teams but also enhances the overall user experience. Through empirical evaluation and comparison with traditional fantasy team drafting methods, we demonstrate the effectiveness of RL in constructing competitive fantasy teams. Our results show that RL-based strategies provide valuable insights into player selection in fantasy sports.

Figures (5)

• Figure 1: Our proposed fantasy team prediction system
• Figure 2: The RL-agent environment interaction process
• Figure 3: Training Progress of PPO Model. The plot shows the cumulative reward over 10,000 episodes of training. The cumulative reward initially starts at a negative value and gradually increases as the training progresses. Eventually, the reward stabilizes, demonstrating the convergence of the PPO model towards optimal behavior.
• Figure 4: This plot illustrates the relationship between the alpha values and the corresponding performance of the model. The shaded area represents the error bounds for each alpha value obtained using the cross-validation datasets
• Figure 5: Density plots showing the distribution of the ratio between the predicted team score and the best team score before and after training across four different cross-validation datasets (a, b, c, and d). In each plot, the solid line represents the distribution of the ratio before training, while the dotted line represents the distribution of the ratio after training. Each plot illustrates how the density shifts towards higher ratios post-training, indicating an improved performance of the agent in selecting teams with scores closer to the best possible team score