Philipp Kahr and Wei Wang explore the effectiveness of machine learning models in detecting anomalies in online chess games, specifically focusing on the Ponziani opening on the Lichess platform. They compare models using absolute values versus percentages, highlighting the limitations of total value-based models, which can be skewed by the overall trend in games played. By calculating the percentage of games featuring the Ponziani opening, they demonstrate improved anomaly detection, offering insights into significant events like the surge in games during the COVID-19 pandemic and the influence of popular chess content creators. The authors detail the process of configuring these models using Kibana Machine Learning, emphasizing the benefits of using percentages for more accurate anomaly detection. They conclude by encouraging readers to explore Elastic Cloud for implementing similar machine learning tasks and suggest exploring related posts in their chess series for further insights.