Progger, a kernel-spaced cloud data provenance logger which provides fine-grained data activity records, was recently developed to empower cloud stakeholders to trace data life cycles within and across clouds. Progger logs have the potential to allow analysts to infer user actions and create a data-centric behaviour history in a cloud computing environment. However, the Progger logs are complex and noisy and therefore, currently this potential can not be met. This paper proposes a statistical approach to efficiently infer the user actions from the Progger logs. Inferring logs which capture activities at kernel-level granularity is not a straightforward endeavour. This paper overcomes this challenge through an approach which shows a high level of accuracy. The key aspects of this approach are identifying the data preprocessing steps and attribute selection. We then use four standard classification models and identify the model which provides the most accurate inference on user actions. To our best knowledge, this is the first work of its kind. We also discuss a number of possible extensions to this work. Possible future applications include the ability to predict an anomalous security activity before it occurs.