Ozone pollution is among the most harmful pollutants negatively affecting human health. Monitoring abnormal ozone levels concentrations is essential to achieve acceptable air quality. This study introduces a data-driven approach for the early detection of abnormal ozone levels in correlated multivariate data. Importantly, the proposed monitoring scheme combined Restricted Boltzmann Machine (RBM) as a features extractor and a one-class support vector machine (OCSVM) as an anomaly detector. The RBM is considered to learn pertinent information without assumptions on the data distribution, and the OCSVM scheme is employed to discriminate normal ozone data from anomalies. The proposed RBM-OCSVM outperformed the standalone OCSVM scheme when applied to actual multivariate ozone data from France. Results revealed the satisfactory performance of the proposed detector in identifying anomalies in ozone measurements.