<p>This study presents a vibration data-driven machine learning approach for accurate fatigue life assessment of suspension coil springs considering the cycle sequences of random loading. This approach integrated wavelet-based signal processing with a neuro-fuzzy modelling framework. Conventional strain-life approaches have inadequate consideration of load sequence effects, which are critical under real-world variable-amplitude loading conditions. To address these challenges, vibration signals collected from road tests were analysed using wavelet transforms to extract time–frequency features, including low-frequency spectral energy and multifractal characteristics. In parallel, strain measurements were used to calculate fatigue damage using the effective strain damage (ESD) model, which captures fatigue cycle sequence sensitivity more accurately than traditional methods. Training with an adaptive neuro-fuzzy inference system (ANFIS), a predictive mapping between the wavelet-derived features and corresponding fatigue lives obtained using the ESD model was developed. Findings support that the wavelet-based vibration spectral energy and multifractality are closely correlated with road conditions and the durability of coil springs. The loading histories under rural road condition which exhibited the highest spectral energy and multifractality due to the rough surface profile obtained the lowest fatigue life (1.121 × 10⁴ blocks to failure). Notably, the proposed ESD–ANFIS hybrid model outperformed the established Morrow-ANFIS model, achieving a Pearson correlation coefficient of 0.957 and significantly reduced root-mean-square error of 0.19. This demonstrates the potential of combining signal processing and soft computing for improved durability assessment, contributing to an accurate fatigue life prediction even in the absence of direct strain measurements.</p>