<![CDATA[This study developed and compared response surface methodology (RSM) and artificial neural network (ANN) models for predicting cotton yarn unevenness (U%) as a function of spacer size and cradle pressure in ring spinning. A 3×3 full-factorial experimental design was employed, yielding nine treatment combinations with 10 replicates each (90 samples total). Due to the limited dataset size (9 averaged data points), both models were fitted and assessed on the same data, with model performance interpreted accordingly. The RSM second-order polynomial model achieved R² = 0.98 (SSE = 0.0055), with sensitivity analysis identifying an optimal U% of 9.056% at a spacer size of 3.3 mm and cradle pressure of 1.53 N within the experimental domain [3.3–4.4 mm] × [1–2 N]. Notably, the unconstrained RSM optimum (x₁ = 2.638 mm) lies outside this domain, suggesting the experimental range may not fully bracket the true minimum. The ANN model, featuring a single hidden layer with four neurons trained via backpropagation over 10,000 epochs, achieved R² = 0.96 (MSE = 0.001942). Both models captured nonlinear input–output relationships, with RSM offering interpretable polynomial coefficients and ANN providing flexible data-driven mapping. A comparative analysis of predictive performance metrics (R², SSE, MSE) is provided, along with a discussion of model limitations related to dataset size and generalizability. These findings offer practical guidance for data-driven parameter optimization in ring spinning processes.]]>
