Nowadays, the better assessment of low back pain (LBP) is an important challenge, as it is the leading musculoskeletal condition worldwide in terms of years of disability. The objective of this study was to evaluate the relevance of various machine learning (ML) algorithms and Sample Entropy (SampEn), which assesses the complexity of motion variability in identifying the condition of low back pain. Twenty chronic low-back pain (CLBP) patients and 20 healthy non-LBP participants performed 1-min repetitive bending (flexion) and return (extension) trunk movements. Analysis was performed using the time series recorded by three inertial sensors attached to the participants. It was found that SampEn was significantly lower in CLBP patients, indicating a loss of movement complexity due to LBP. Gaussian Naive Bayes ML proved to be the best of the various tested algorithms, achieving 79% accuracy in identifying CLBP patients. Angular velocity of flexion movement was the most discriminative feature in the ML analysis. This study demonstrated that: supervised ML and a complexity assessment of trunk movement variability are useful in the identification of CLBP condition, and that simple kinematic indicators are sensitive to this condition. Therefore, ML could be progressively adopted by clinicians in the assessment of CLBP patients.