Abstract
The worldwide demand for oil has been rising rapidly for many decades, being the first indicator of economic development. Oil is extracted from underneath reservoirs found below land or ocean using oil wells. An offshore oil well is an oil well type where a wellbore is drilled underneath the ocean bed to obtain oil to the surface that demands more stability than other oil wells. The sensors of oil wells generate massive amounts of multivariate time-series data for surveillance engineers to analyze manually and have continuous insight into drilling operations. The manual analysis of data is challenging and time-consuming. Additionally, it can lead to several faulty events that could increase costs and production losses since the engineers tend to focus on the analysis rather than detecting the faulty events. Recently, machine learning (ML) techniques have significantly solved enormous real-time data anomaly problems by decreasing the data engineers' interaction processes. Accordingly, this study aimed to utilize ML techniques to reduce the time spent manually to establish rules that detect abnormalities in oil wells, leading to rapid and more precise detection. Four ML algorithms were utilized, including random forest (RF), logistic regression (LR), k-nearest neighbor (K-NN), and decision tree (DT). The dataset used in this study suffers from the class imbalance issue; therefore, experiments were conducted using the original and sampled datasets. The empirical results demonstrated promising outcomes, where RF achieved the highest accuracy, recall, precision, F1-score, and AUC of 99.60%, 99.64%, 99.91%, 99.77%, and 1.00, respectively, using the sampled data, and 99.84%, 99.91%, 99.91%, 99.91%, and 1.00, respectively, using the original data. Besides, the study employed Explainable Artificial Intelligence (XAI) to enable surveillance engineers to interpret black box models to understand the causes of abnormalities. The proposed models can be used to successfully identify anomalous events in the oil wells.