Monitoring covariance in multivariate time series: Comparing machine learning and statistical approaches