MANUSCRIPT Mapping Rare Earth Element Prospectivity in the Ruri Carbonatite Complex Using Explainable Ensemble Learning

Imagine you are a treasure hunter, but instead of gold or jewels, you are looking for the invisible ingredients that make your smartphone work and your electric car zoom. These ingredients are called Rare Earth Elements (REEs). They are the "spices" of modern technology, essential for everything from wind turbines to medical devices. The problem is, finding them is like trying to find a specific needle in a haystack that is constantly moving and changing shape.

To find these needles, geologists used to rely on digging holes and guessing based on what rocks looked like nearby. But today, we have a new superpower: computers that can learn. Think of machine learning as a very smart student that can read thousands of maps and chemical reports to spot patterns humans might miss. However, there's a catch. These smart computers are often "black boxes." You ask them, "Where is the treasure?" and they say, "Here," but they can't explain why. It's like a friend who gives you directions but refuses to tell you which landmarks to look for. This makes geologists nervous because if they don't understand the "why," they can't trust the "where."

This is where a new approach called "Explainable Artificial Intelligence" (XAI) comes in. It's like giving the computer a voice so it can say, "I think the treasure is here because the rocks are a certain color, the ground is cracked in a specific way, and the chemicals smell like this." By combining the brainpower of many different computer models into one "super-team," scientists hope to not only find the treasure but also understand the map perfectly.

---

The Story of the Ruri Hills Treasure Hunt

In the southwest of Kenya, there is a place called the Ruri Carbonatite Complex. It's a rocky landscape where ancient magma pushed its way up from deep underground, leaving behind a complex mix of rocks that might be hiding a stash of Rare Earth Elements. But the area is under-explored, meaning we don't have a complete map of where the good stuff is. A team of researchers from the University of Nairobi decided to tackle this mystery using a high-tech detective method called "Stacked Ensemble Learning."

Think of their method like a panel of expert judges at a talent show. Instead of relying on just one judge, they brought in three different types of "judges" (computer algorithms) to look at the data:

1. The SVM (Support Vector Machine): A judge that is great at drawing clear lines between different groups of rocks, even when the groups are messy.
2. The RF (Random Forest): A judge that builds hundreds of tiny decision trees to make sure it doesn't get fooled by a single weird rock.
3. The XGBoost: A judge that learns from its mistakes, getting smarter with every step it takes.

Usually, you might pick the one judge who gets the most answers right. But this team did something clever: they created a "Head Judge" (a meta-learner) who listens to all three of the other judges and combines their opinions to make the final decision. This is the "stacked" part. It's like asking three different experts for advice and then having a wise mentor weigh all their answers to give you the best possible recommendation.

The team fed this super-team a massive amount of data: chemical concentrations of elements like Thorium and Uranium, the types of rocks present, and the location of cracks and faults in the ground. They taught the model to distinguish between rocks that are rich in Rare Earths and those that are not.

What They Found

The results were impressive. When they tested their "super-judge" panel, it got the answer right 95.6% of the time. It was so good at spotting the right spots that its "score" for separating good from bad areas was 0.97 (on a scale where 1.0 is perfect). Interestingly, the single best judge on the team (XGBoost) was already doing a great job on its own, but the team approach made the predictions even more stable and reliable.

But the real magic wasn't just the score; it was the explanation. Using a tool called SHAP (which acts like a magnifying glass to see what the computer is thinking), the researchers discovered exactly what clues mattered most. The model told them, loud and clear, that the most important signs of a treasure trove were:

• Total Rare Earth Oxides (TREO): The overall amount of the "spices" in the rock.
• LREE/HREE fractionation: A specific ratio of two types of Rare Earths that acts like a fingerprint for this kind of mineral.
• Thorium and Uranium: These elements often hang out with Rare Earths, so finding them is a big clue.
• Structural density: How many cracks and faults are in the area, which act as highways for the mineral-rich fluids.

The final map they produced highlighted specific zones in the Ruri Complex where the chances of finding these elements are very high. These zones are linked to the carbonatite rocks, the altered rocks around them (called fenites), and the main fault lines.

Why This Matters

The paper suggests that this method is a game-changer because it doesn't just give a "black box" answer. It gives a clear, data-driven map that geologists can trust because they can see why the computer pointed to a specific spot. The researchers found that the best spots are in the central parts of the complex, right where the rocks and chemical signals line up.

While the model is very accurate, the authors are careful to note that this is a guide, not a guarantee. They suggest that these high-probability zones should be the first places for real-world fieldwork, like digging trenches or drilling holes, to confirm the findings. By combining the power of a team of smart computers with the ability to explain their reasoning, this study offers a new, transparent way to hunt for the critical minerals our future depends on.