<p>The surface strengthening process has gained increasing popularity in both scientific research and engineering applications due to its effectiveness in enhancing fatigue resistance and ensuring the long service life of critical engineering components. However, conventional life prediction models often fail to capture the influence of complex microstructural features and intrinsic mechanical mechanisms, limiting their predictive reliability. This study proposes a preprocessing neural network model for accurately predicting the fatigue life of surface-strengthened Ti–6Al–4V alloys. The model integrates surface integrity features, including surface roughness, residual stress, and hardness, alongside fatigue test parameters. Both residual stress and hardness exhibit continuous depth-dependent gradients from the surface to the core. To address the challenges posed by such correlated and continuously varying inputs, the network architecture incorporates an internal preprocessing module that enables logical feature dimension reduction while preserving essential physical information. The proposed neural network model demonstrated robust predictive accuracy across an extensive dataset of surface-strengthened Ti–6Al–4V, outperforming existing machine learning models trained with features derived from PCA-based dimensionality reduction. Additionally, a random forest model was employed to evaluate and rank the contribution of individual features to fatigue life. This analysis underscores the relative importance of surface strengthening factors and offers valuable insights for optimizing fatigue-resistant design strategies.</p>