As we know from medicine, prevention is better than cure. To avoid future problems, we have to recognize them earlier; therefore, we need prediction. Because machine learning algorithms have the potential to make more accurate predictions, many scientists and researchers, etc. have already started using them. Although these machine learning algorithms are used to create predictions, they also look for patterns within the value labels assigned to data points. There are two main types of machine learning, supervised learning and unsupervised learning. In addition, there are also the so-called semi-supervised learning methods, which are a combination of the two main learning methods. In this thesis, the hidden Markov model (HMM), an unsupervised learning method, is used to analyze time-series data and find the hidden states, which can be used for predicting problems that may arise in oil fields, especially in petroleum production, e.g., sucker rod pump failure diagnosis. This thesis starts with the basics and theory of HMM. Then the three main problems of HMM and the solution for the problems will be discussed. Moreover, the tools and programming languages available to generate our own algorithms and functions required for the model will be discussed. Then hidden Markov model will be used to find the start and the end of up-and downstrokes from the dataset. Finally, using HMM to observe the sucker rod pump operation over time (finding hidden states), first for the entire dataset and then for a selected part of the dataset. The results from the hidden Markov model will be compared with other clustering methods, namely the Gaussian Mixture and K-Means.