drillinganomalymetrics 0.1.0

Package providing functions to compute a modified recall, precision and F1-score metrics for drilling anomaly detection models evaluated on historical data sets.

drillinganomalymetrics

---

Description
drillinganomalymetrics is a Python package containing functions to compute accuracy metrics in the context of drilling anomaly detection. These metrics are design to reflect the proximity of a drilling anomaly prediction model to its ideal characteristics, namely:

• It effectively detects early signatures of all stuck pipe events without missing any,
• It effectively detects sticking symptoms as early as possible, allowing the drilling team to take actions to prevent risk escalation, and
• It does not generate false warnings, i.e., predictions of high risk in non-risky conditions.

Why New Metrics
In summary, traditional classification metrics fail to reflect how close a model is to the aforementioned ideal characteristics. More importantly, they often generate misleading values: bad models can get high scores and good models very low values. Moreoever, traditional metrics can become artificially inflated when used in a point-wise manner in drilling anomalies that develop over long periods of time. Examples are provided in Montes et al. (2025)—SPE-220725-PA.

---

Associated Theory/Methodology
The metrics provided in this package are thoroughly described and illustrated through examples in the context of stuck pipe prediction in Appendix B in the following open-access SPE Journal article:

• Montes, A. C., Ashok, P., and van Oort, E. 2025. Review of Stuck Pipe Prediction Methods and Future Directions. SPE Journal 30(06): 3334–3363. SPE-220725-PA. Link.

---

---

Abraham C. Montes

LinkedIn | Google Scholar | Personal Website