<p>Addressing the challenge of accurately measuring and predicting spindle rotational errors during CNC machine tool cutting operations, this paper proposes an effective error separation method and establishes a high-precision prediction model. This provides a basis for implementing error compensation and enhancing machining accuracy. A spindle rotational error measurement model is established using the three-point method. The Whale Optimisation Algorithm (WOA) is employed to optimise sensor mounting angles, thereby eliminating harmonic suppression. An experimental platform is constructed to collect displacement and vibration signals under varying machining parameters, with spindle rotational errors extracted via error separation techniques. Furthermore, the Hybrid Strategy Flower Pollination Algorithm (HSFPA) is integrated to optimise the Radial Basis Function (RBF) network, constructing the HSFPA-RBF prediction model. Experimental results indicate that optimising sensor installation angles to <i>α</i> = 15.47° and <i>β</i> = 31.64° effectively prevents harmonic suppression. The established HSFPA-RBF model demonstrates optimal predictive performance, achieving a root mean square error (<i>RMSE</i>) of 5&#xa0;μm, a mean absolute error (<i>MAE</i>) of 3.4&#xa0;μm, and a coefficient of determination (<i>R</i>²) of 0.9442. This methodology effectively captures spindle rotational errors during cutting operations. The developed model exhibits high predictive accuracy, providing technical support for online error compensation and process optimisation.</p>