Machine learning is increasingly being applied to petrophysical and geological problems. However, building these models and comparing the results can be a long drawn out manual process. By testing multiple combinations of input curves, feature selection methods and prediction algorithms this process can turn into dozens of modelling runs, which can take hours when carried out manually.
The real issue is not only about speed. Manual predictive workflows often rely on trial and error, which can result in inconsistent results and poor optimisation. Additionally, manual processing can allow human error and biases to creep in.
When results are needed to make fast and robust decisions, automation of the predictive workflow, whilst maintaining transparency becomes more of a necessity.
When manually building predictive models and workflows, identifying the correct inputs (also known as features) can become a tedious guessing game.
One approach some take is to add every possible input into the model. This can lead to models that are bloated with unpredictable performance. When all available inputs are used in a model, it starts to become difficult to interpret and understand how the model has reached it’s conclusion based on the inputs. If some of those inputs are unavailable in another well, it can make it difficult to compare results across wells or scale the model to new wells within the project.
Feature Selection is an essential step in any machine learning modelling process. By including the wrong inputs in the process, we can add complexity without adding any real value. If we remove irrelevant inputs, we can reduce the noise in the model, improve accuracy and shorten model training times.
By leveraging automation, we can introduce structure and repeatability into the modelling process, whilst maintaining trustworthiness and reducing human bias and error. However, this does not mean we are handing the whole process over to a machine. The user is still in control and can visualise and interpret the whole process.
IP 2023 saw the introduction of Experienced Eye, a powerful tool for Feature and Predictive model selection for regression based problems. This module provides a way to bring everything together for your machine learning predictive workflow for curve prediction tasks such as:
This made it convenient and fast whilst maintaining transparent and repeatable workflows in a centralised module. This allowed us to see what combinations of input curves work best, and understand the impacts of the different combinations of inputs. This helped us to make informed decisions without second guessing the choices and the models.
Now with our latest update to IP 2025 (Update 3), the Experienced Eye module has received a powerful upgrade in speed, is multi-well aware and introduces new powerful data visualisations.
There are numerous improvements to the Experienced Eye Module that make it easier and much smoother to work with data from multiple wells.
Some of these include:
A complete rework of the input selection, which closely mirrors that of other Curve Prediction modules in IP. This brings familiarity whilst adding extra capabilities including cell highlighting when curves are missing from a selected well and choosing whether wells can be used for training, testing or both.
An updated results view with powerful visualisations to help you identify the best performing models and input combination. Improved charts now allow you visualise how different wells perform relative to each other, how prediction results vary well by well, how different combinations of input curves effects the final model, and much more.
In addition to introducing multi-well capabilities within the module, we have also improved the log plots, histograms and crossplots so that they are full interactive allowing you to full explore your prediction results.
Once you have chosen your final model, you can quickly and easily export it to IP’s standalone Curve Prediction models and build upon it as new wells become available.
The adoption of machine learning in petrophysics is not just about speed, it is about making petrophysical machine learning modelling practical, scalable and interpretable whilst maintaining transparency and user control.
Experienced Eye centralises the feature selection and model comparison for curve prediction problems, such as predicting incomplete logging curves and repairing poor quality data. This makes the process of comparing different models and inputs smoother, faster and more convenient whilst allowing you to understand how the models are built and perform.
With the latest release (IP 2025.3), Experienced Eye has evolved further by delivering significant speed improvements, multi-well capability, and powerful new visualisations. This means that you can now compare feature rankings, evaluate model performance, and make informed decisions across entire projects in a fraction of the time, without sacrificing control or interpretability.
Arkalgud, R., McDonald, A. and Brackenridge, R., 2021. Automated Selection of Inputs for Log Prediction Models Using a New Feature Selection Method. SPWLA 62nd Annual Logging Symposium. [online] Virtual Event, May. Available at: https://doi.org/10.30632/SPWLA-2021-0091
Arkalgud, R., McDonald, A. and Brackenridge, R., 2020. Automated Selection of Inputs for Log Prediction Models Using Domain Transfer Analysis DTA Derivative. Abu Dhabi International Petroleum Exhibition & Conference. [online] Abu Dhabi, UAE, November. Available at: https://doi.org/10.2118/203094-MS
Arkalgud, R., McDonald, A. and Brackenridge, R., 2021. Automated Selection of Inputs for Log Prediction Models Using a New Feature Selection Method. SPWLA 62nd Annual Logging Symposium. [online] Virtual Event, May. Available at: https://doi.org/10.30632/SPWLA-2021-0091
Arkalgud, R., McDonald, A. and Crombie, D., 2019. Domain Transfer Analysis – A Robust New Method for Petrophysical Analysis. SPWLA 60th Annual Logging Symposium. [online] The Woodlands, Texas, USA, June. Available at: https://doi.org/10.30632/T60ALS-2019_HHHH
McDonald, A., 2021. Data Quality Considerations for Petrophysical Machine-Learning Models. Petrophysics, [online] 62, pp.585-613. Available at: https://doi.org/10.30632/PJV62N6-2021a1