Detection of anomalous behaviour in a network is crucial to ensure the safety of data. However, developing an effective anomaly detection mechanism is difficult. This is because these mechanisms have a comparatively higher tendency to raise false alarms. Thus, development of effective training mechanisms for predictive models such as machine learning and deep learning is of paramount importance. At the same time, recent datasets must be considered in order to train the predictive models. In this paper, we have considered recent datasets, tested them in balanced and unbalanced formats over various machine learning and deep learning models. In our analysis, we found that it is crucial to consider imbalanced data in order to generate an efficient anomaly detection model, even if it trained on balanced datasets.