Electrical submersible pumps (ESP) are a key artificial lift method, particularly in challenging oilfield operations involving highly viscous fluids such as heavy oil, emulsions, or subsea boosting applications. Traditional empirical models for predicting ESP performance often face significant limitations, including substantial deviations due to reliance on restricted datasets. This study presents a novel prediction model addressing these challenges by leveraging a comprehensive experimental dataset encompassing diverse operating conditions, fluid viscosities, and impeller geometries. The model predicts single-stage performance curves and extends applicability to multi-stage ESPs through isothermal and non-isothermal approaches, incorporating energy dissipation effects. Validation against experimental data demonstrates accurate predictions for head, flow rate, and efficiency, significantly reducing errors compared to existing methods. Additionally, the prediction model and multi-stage approaches were tested with datasets from other authors. Even when applied to different ESP geometries, the model demonstrated good accuracy in estimating correction factors and predicting complete performance curves.