Yes—machine learning can be used to predict whether a person has a formal financial account from recorded characteristics, but the result is a prediction, not proof of why the person owns an account or what intervention would change that. In this introductory example, account ownership is the yes-or-no target; factors such as age, education, income, employment, location, phone ownership, and internet access are possible features, not confirmed inputs from a particular dataset.
What does “financial inclusion” mean in this example?
The example turns one measurable outcome—whether an individual has a formal financial account—into a binary classification problem. The account indicator is the target the model is asked to predict; the recorded characteristics supplied to it are features.
As an Amazon Associate I earn from qualifying purchases.
That outcome is useful, but narrower than financial inclusion as a whole. The World Bank counts accounts at banks and regulated institutions, including credit unions, microfinance institutions, and mobile-money providers. Its Global Findex 2021 summary calls account ownership “the fundamental measure of financial inclusion and the gateway to using financial services in a way that facilitates development.” Ownership alone, however, does not establish that services are affordable, suitable, accessible in practice, actively used, or improving someone’s financial well-being.
What do the available figures say—and which year do they describe?
The World Bank’s Global Findex 2021 reported that 76 percent of adults worldwide had an account in 2021, up from 51 percent in 2011. In developing economies, 71 percent had an account in 2021, compared with 63 percent in 2017. The gender gap in account ownership in developing economies was 6 percentage points in 2021, down from 9 percentage points. These are dated survey findings, not current 2026 rates.
#1 Best Overall
The 2021 edition drew on nationally representative surveys of almost 145,000 people in 139 economies, representing 97 percent of the world’s population. The World Bank’s newer Global Findex 2025 edition is based on surveys of about 148,000 adults in 141 economies conducted during calendar year 2024. Its download page lists country, regional, and income-group indicators for 2024, 2021, 2017, 2014, and 2011, across topics including accounts, payments, savings, credit, resilience, phone ownership, internet use, and digital safety. An aggregated indicator series should not be assumed to be individual-level records or directly interchangeable with another measure.
For example, the World Bank reported that 35 percent of unbanked adults in Sub-Saharan Africa cited not having a mobile phone as a reason for lacking a mobile-money account. This is a reported barrier, not a model result or an estimate of the causal effect of providing a phone.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
How would an account-ownership model be built?
1. Define the prediction target and timing
Specify exactly what counts as an account, which population and location are in scope, and when the prediction is meant to be made. The definition should match the survey or administrative source. Decide which class is “positive”—for example, account ownership—and what action, if any, would follow a prediction.
2. Inspect the data before choosing features
Potential inputs in the introductory example include age, education, employment, income, location, phone ownership, internet access, and gender. They are illustrative possibilities, not a guarantee that a specific dataset contains them or that each is appropriate to use. Check variable definitions, missing values, survey year, geography, sampling design, and access conditions. The World Bank’s 2021 Findex catalog describes that edition as public and nationally representative; it may be a starting point, but its documentation must be checked before analysis. The original tutorial does not identify a dataset it actually used, so it should not be described as a Findex analysis.
3. Prepare records without leaking the answer
Clean and explore the records, handle missingness transparently, and encode categorical values in a way the selected model can use. Most importantly, include only information that would genuinely be available at the prediction time. A field recorded after account opening, or one that directly encodes account ownership, can let a model appear accurate by revealing the answer rather than learning a useful pattern.
4. Split, fit, and evaluate for the intended setting
Separate data used to fit the model from data used to evaluate it. The split should reflect the sampling and deployment context; a random split does not by itself make an evaluation representative of future use or eliminate leakage. Possible classifier families named in the tutorial include logistic regression, decision trees, random forests, gradient boosting, support-vector machines, and neural networks. They are modeling choices, not a ranking: no comparative results are reported to establish a best-performing approach.
Rank #4
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
How should model performance be judged?
Do not rely on a single accuracy figure. If account ownership is much more common than non-ownership, a model that mostly predicts the common class can have high accuracy while failing to identify people without accounts. The tutorial’s 80/20 split and 85-percent accuracy are illustrative examples, not measured results.
Recommended Free Tools
Useful measures include accuracy, precision, recall, F1 score, ROC-AUC, and a confusion matrix. Choose them according to the purpose and the consequences of errors. Precision asks how often predicted positives are actually positive; recall asks how many actual positives the model finds. For an intervention aimed at reaching unbanked people, the cost of missing someone without an account may differ from the cost of mistakenly flagging an account holder. Define the positive class and decision threshold explicitly, and examine calibration and relevant subgroup errors as well as overall scores.
Best Value
What a prediction can—and cannot—tell you
A feature associated with account ownership may help predict it without causing it. If phone access, income, employment, or location is predictive, that does not show that changing the feature would produce account ownership. As the tutorial puts it, “Prediction does not automatically establish causation.” Causal claims require a study design and assumptions capable of supporting them.
Reported barriers can suggest questions for follow-up work, but a classifier cannot establish that a particular policy will remove them. The World Bank’s account-ownership summary describes lack of money, distance to a financial institution, and insufficient documentation among commonly cited reasons for not having an account. Establishing which response would help requires evidence beyond predictive associations.
What responsible use requires
Before using predictions in decisions, examine who is represented in the records and who is missing; whether account definitions or error rates differ across groups; and whether sensitive attributes or their proxies are being used. Protect personal data, explain the model’s role and limits, and retain meaningful human oversight where outputs affect people. These checks are starting points, not a complete governance standard or legal opinion. The introductory example does not specify a jurisdiction, legal requirement, fairness metric, or validated governance protocol.
For anyone building an individual-level analysis, inspect the relevant microdata release and its documentation rather than assuming a public download of aggregate indicators contains suitable person-level rows. A credible result depends on the source, outcome definition, timing, data quality, evaluation design, and intended use—not simply on selecting an algorithm.
Read the introductory tutorial on DEV Community. For context and data documentation, see the World Bank’s Global Findex 2021 account-ownership summary, the 2021 Global Findex Data Catalog, and the Global Findex download-data page.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
