Fresh Thinking: The 3-Minute Read. Episode 101

Busy day? Short on time?
Here’s your 3-minute quick read version of Fresh Thinking by Snowden Optiro’s podcast Episode 101.
We’ve pulled out the key insights and takeaways—perfect for when you want the value, minus the headphones.

Fresh Thinking – 3-minute Read of Episode 101:
Machine Learning for Geological Domaining Case Study

In Episode 101 of Fresh Thinking by Snowden Optiro, Managing Consultant Susan Havlin and Dr Gregory Zhang spoke about how machine learning – specifically the K-means algorithm – can enhance geological domaining in mineral resource estimation.

Here are the key points and insights from the podcast.

The Challenge:

Traditional geological domaining relies heavily on manual interpretation of geological logging and grade data, using explicit or implicit modelling techniques. However, in this case study of a gold deposit, these methods led to irregular, geologically inconsistent domains, posing risks to model accuracy and downstream processes.

With over 30,000 samples comprising both numeric and categorical data, the team sought an alternative approach to better handle the dataset’s complexity.

Why K-means?

K-means is an unsupervised machine learning algorithm that clusters data into distinct groups based on similarities. In this application, it was used to differentiate ore and waste regions by clustering similar geological and grade data points together.

Key reasons it was selected:

• Handles large, complex datasets efficiently.
• Identifies patterns not immediately obvious through manual interpretation.
• Provides a quick, initial separation of data into potential ore and waste domains.

How It Worked:

1. Preprocessing: Normalised numeric features and converted categorical data (like mineralogy) into numeric values.
2. Model Training: Determined the appropriate number of clusters and ran the K-means algorithm, experimenting with various parameters.
3. Validation: Visualised clusters in 3D to assess geological coherence, followed by additional iterations and adjustments based on geological expertise.

Key Insights:

• Speed & Efficiency: K-means processed the dataset quickly and highlighted patterns missed by traditional methods.
• Refinement Required: Some cluster boundaries lacked geological coherence, so domain refinement with human geological knowledge was essential.
• Complementary Tool: K-means is not a replacement for geological judgement—it’s a tool to assist, particularly useful when working with large datasets.
• Accessible to Geologists: With basic programming knowledge or embedded software tools, resource geologists can easily integrate K-means into their workflow.

Final Thought:

Machine learning, when thoughtfully applied, adds a valuable extra layer to geological domaining – enhancing, not replacing, the geologist’s expertise.

---

Interested in applying advanced techniques to your resource models? Contact Dr Gregory Zhang: contact@snowdenoptiro.com

---

---

Listen to or Watch the Full Episode

Prefer to listen to the audio version on this podcast?
You can watch/listen to the full episode of this Fresh Thinking by Snowden Optiro podcast on YouTube, Spotify, Apple Podcasts, Amazon Music and most other podcast platforms.

Prefer to watch the video version of this podcast?
Here is a link to our Snowden Optiro YouTube channel: https://youtu.be/gEqWM6z0kwI?si=foDXiZP5K4llGQLw

LINK to Snowden Optiro website