This thesis aimed to find unknown sucker rod classification parameters next to the current parameter “load cycles” as this was already known from literature and is currently utilized as the solitary classification parameter at OMV. Additionally, the question was prompted whether other global oil companies rely on alternative classification parameters. The target was to present newly investigated classification parameters to classify sucker rods in order to optimize their application and cut down the sucker rod failures causing dead periods. By gathering the data of the sucker rod failure cases that occurred from 2016 to 2019 at OMV, a dataset was created. The dataset was composed of 26 failure and wellbore parameters, predefined by OMV and Montan University Leoben. The determination of new classification parameters was performed by visualization methods, unsupervised machine learning methods and supervised machine learning methods. Scatter plots were utilized as the visualization method, to identify possible relations in between the contrasted parameters. The dataset was later clustered with unsupervised learning methods, while the normed distance values expressed the clustering quality and indicated how effective the data separation by certain parameters was. Artificial neural networks were also created with the aid of supervised machine learning and were optimized by the sequential forward selection method to determine the most important parameters for sucker rod failure. Visualization methods revealed that corrosion is the dominating failure reason for sucker rods in comparison to break at a ratio of roughly 60 to 40. Especially the protector region represents an accumulation point for sucker rod failure, supposedly because of the insufficient corrosion inspection beneath the sucker rod guides. It strikes out that only a small amount of the failed “new” sucker rod strings has reached the predefined load cycles of 16 million. It is recommended to apply the standard corrosion measurement by corrosion coupons to the wells with sucker rod failures, to detect the corrosion problems more properly. The execution of the unsupervised leaning methods PCA, k-mean clustering and hierarchical clustering was also successful. The gained results indicated slight relations within the dataset, which are presumably clarified better by an extension of the data set. The application of supervised machine learning was provided reliable results. Recognizable was a strong relation between “failure position” and “feature1”, so the failure position on the rod is dependent on the total load applied to the rod over the whole lifetime. However, none of the achieved results allowed the extraction of new classification parameters. The reason is most probably the minor size of the data set, whereby the clarity and significance of the results are decreased. Therefore, tagging of each individual sucker rod is endorsed in order to track a reliable value for the number of “load cycles” for used sucker rods and extend the current data. The market research performed for alternative sucker rod classification parameters was supported by the statements of multiple global oil companies but didn’t deliver new findings. The successful execution of the presented methods proves the correct choice of investigation methods. The combination of the presented methods provides a solid interpretation base. A dataset extension followed by a re-application of the presented methods should deliver more precise results.