The oil and gas industry, like many others, is facing challenges brought about by the energy transition, demanding the optimization of operations within defined boundaries. In terms of geo-energy exploration related activities, this is associated with a cost-effective and safe drilling operation. In this regard, it is crucial to minimize the occurrence of undesired downhole problems, which may delay the drilling process, potentially causing non-productive time. One of the essential keys to achieving that is the early detection of anomalous downhole behaviour by continuously monitoring the surface-measured drilling parameters. The hydraulic system, along with other key surface parameters, plays a crucial role in successful drilling operations. It not only facilitates the circulation of drilling fluids, hole cleaning, and bit colling, but also provides valuable insights of the current downhole condition. Accurate modelling and monitoring of the surface-measured standpipe pressure can serve as a reliable indicator of potential anomalous downhole behaviour. However, the conventional physics-based approach for modelling standpipe pressure faces limitation in accurately representing the dynamic and complex nature of the downhole condition. Regarding this issue, the ultimate goal of this thesis is to develop a data-driven model based on a machine learning concept to predict standpipe pressure with only three controllable surface parameters as input for the model and still provide robust estimates of the target variable. The models are trained with trouble-free drilling data, which should allow the model to represent the normal trend and thus provide means for analysis and trend identification by comparison of the actual value with the modelled values. In conclusion, the applied methodology and algorithm can provide acceptable estimates of the target variable utilizing minimum required datapoints stemming from the same well. However, an optimization of the applied approach can possibly lead to improved results. The provided confidence interval provides a range of possible values for the target variable, thus can be useful for analysis and real-time monitoring. However, the predicted confidence interval cannot be directly interpreted as a safe operation window.