Stacking Ensemble Models for Predicting Digital Payment Adoption: A Comprehensive Machine Learning Analysis of Global Findex 2025 Data

Imagine the global economy as a massive, bustling marketplace where everyone is trying to trade, save, and buy things. For a long time, this marketplace was mostly physical—you needed to walk to a bank, carry cash, or write a check. But recently, a magical new tool has appeared: digital payments. These are like invisible coins that zip through the air via smartphones and computers, letting people pay for things without ever touching paper money. This shift is a big deal because it helps people who were previously locked out of the market—like those in remote villages or without much money—to finally join the party. This is called "financial inclusion," and it's the key to helping everyone build a better life.

But here's the tricky part: not everyone is using these digital coins yet. Some people are jumping in with both feet, while others are standing on the sidelines, confused or scared. Scientists and economists have been trying to figure out exactly why some people adopt these new tools while others don't. They look at clues like how much money someone makes, how old they are, and whether they have a smartphone. Traditionally, they used simple math to guess the answer, but the world is messy and complicated. Sometimes, the clues interact in weird ways—like how having a smartphone doesn't help if you don't know how to use it, or how having money doesn't help if you don't trust the internet. To solve this puzzle, researchers are now using "machine learning," which is like teaching a super-smart robot to look at millions of clues at once and find patterns that human brains might miss.

In this study, a team of researchers decided to build a super-robot detective to solve the mystery of digital payment adoption. They didn't just ask one robot; they built a whole team of them, each with a different way of thinking, and then taught them to work together. They used a massive dataset called the "Global Findex 2025," which is like a giant photo album containing information about 144,090 adults from 141 different countries. After cleaning up the messy parts of the data, they had a clear picture of 102,954 people to study.

The researchers started by giving their robot team a list of 183 different clues (variables) to look at, such as education levels, internet access, and income sources. To make the job easier, they used a smart filter to pick out the 66 most important clues and then combined some of them into new, super-clues. For example, they created a "super-clue" that multiplied a person's education level by their internet usage, because the team suspected that having both together was much more powerful than having just one. They also made new clues about how many different ways a person earns money and how "financially inclined" they are (like whether they save money or have a credit card).

Next, the researchers trained three different types of "base" robots using advanced algorithms called XGBoost, LightGBM, and CatBoost. Think of these as three different detectives: one is great at spotting small details, another is fast at processing huge amounts of information, and the third is excellent at handling messy, categorical data. They didn't just let these robots guess; they used a special tool called Optuna to fine-tune their settings, like adjusting the focus on a camera lens, to make sure they were as sharp as possible.

But the real magic happened when they decided to put these robots into a "stacking" team. Instead of letting one robot make the final decision, they had the three base robots make their guesses first. Then, they fed those guesses into a "meta-learner" robot (in this case, an ExtraTrees classifier) whose only job was to listen to the three base robots and decide who was right. It's like having three experts give their opinions, and then a wise judge listens to all of them to make the final verdict. The researchers tested this team against other methods, like just asking one robot or having the robots vote by majority.

The results were impressive. The "stacking" team, specifically the combination of LightGBM and XGBoost as the base detectives with the ExtraTrees robot as the judge, got it right 93.04% of the time. This was the best score they could find, beating out the single robots and other team combinations. The researchers also used a tool called SHAP to peek inside the robot's brain and see why it made its decisions. They found that the robot was heavily influenced by three main factors: how much education a person had, whether they had access to the internet, and how much money they made. Interestingly, the robot learned that education and internet access work best when they happen together; having one without the other wasn't as helpful.

The study also showed that while fine-tuning the robots (hyperparameter optimization) helped a little, the real secret sauce was the team structure itself. The "stacking" approach made the predictions much more stable and accurate than any single robot could achieve alone. However, the researchers noted that their robot still struggled a bit with predicting the behavior of low-income and rural populations. This suggests that for these groups, there are still hidden barriers—like lack of trust, security fears, or slow internet speeds—that the robot couldn't fully explain with just the data it had.

In the end, this paper shows that by using a team of smart, specialized machine learning models working together, we can predict who will start using digital payments with very high accuracy. It confirms that education and internet access are the golden keys to unlocking financial inclusion, but it also warns us that simply building the infrastructure isn't enough; we need to make sure people have the skills and trust to use it. The researchers suggest that future studies should look at how these behaviors change over time and include bigger economic factors, but for now, this "stacked" team of robots gives us our clearest view yet of how the digital money revolution is spreading across the globe.