Overfitting (Over-Optimization)

Overfitting – When a robot’s parameters are too finely tuned to historical data noise, resulting in poor future performance. Also called curve fitting. Signs include excellent in-sample results but large performance drops on fresh data. Avoiding overfitting involves techniques like limiting parameter count, forward testing, or cross-validation. As one MQL5 blog explains, adding many parameters exponentially increases combinations and can lead to “some good output results [that] can sometimes be random,” making the system lack robustness. Forward/backtesting on separate data segments (see Forward Testing and Walk-Forward) is a common safeguard.

⬅ Robots Glossary