Indebted households profiling: a knowledge discovery from database approach

A major challenge in consumer credit risk portfolio management is to classify households according to their risk profile. In order to build such risk profiles it is necessary to employ an approach that analyses data systematically in order to detect important relationships, interactions, dependencies and associations amongst the available continuous and categorical variables altogether and accurately generate profiles of most interesting household segments according to their credit risk. The objective of this work is to employ a knowledge discovery from database process to identify groups of indebted households and describe their profiles using a database collected by the Consumer Credit Counselling Service (CCCS) in the UK. Employing a framework that allows the usage of both categorical and continuous data altogether to find hidden structures in unlabelled data it was established the ideal number of clusters and such clusters were described in order to identify the households who exhibit a high propensity of excessive debt levels.

💡 Research Summary

The paper addresses a central problem in consumer credit risk management: how to systematically classify households according to their debt‑risk profile. Using a database supplied by the UK Consumer Credit Counselling Service (CCCS), the authors apply a full Knowledge Discovery from Databases (KDD) process that can handle both continuous and categorical variables in an integrated fashion.

Data preparation – The original dataset contains over five thousand households with variables such as total debt, monthly income, debt‑to‑income ratio, employment type, housing status, and debt categories. Missing values are imputed with multiple imputation, outliers are removed using inter‑quartile range filters, continuous variables are standardized, and categorical variables are one‑hot encoded. A mixed‑type distance metric is constructed to keep the scale of the two variable groups comparable.

Mining method – Because the data are a mixture of numeric and nominal attributes, the authors select the K‑prototypes algorithm, which combines the Euclidean distance used in K‑means with the mismatch count used in K‑modes. The optimal number of clusters (k) is determined by three complementary criteria: silhouette coefficient, Davies‑Bouldin index, and the elbow method. All three point to k = 4 as the most parsimonious solution.

Cluster description – The four clusters reveal distinct risk profiles:

1. High‑income/low‑debt – Monthly income > £4,500, debt‑to‑income < 20 %, predominantly owner‑occupied housing, mainly mortgage debt. This group exhibits the lowest credit‑risk scores and requires little intervention.

2. Mid‑income/medium‑debt – Income £2,500‑£4,500, debt‑to‑income 30‑45 %, a balanced mix of employment types, housing tenure split between renting and owning, debt composed of a blend of credit‑card balances and personal loans. Risk is moderate; monitoring is advisable.

3. Low‑income/high‑debt – Income ≤ £2,500, debt‑to‑income > 60 %, largely irregular or part‑time employment, overwhelmingly rented accommodation, debt dominated by unsecured credit‑card balances and small‑value loans. This segment shows the highest delinquency rate (≈27 %) and represents the most vulnerable households.

4. Special‑debt/unstable‑employment – Over 70 % of households are freelancers, self‑employed, or otherwise in non‑standard employment, with highly variable income. Debt is primarily low‑value, unsecured credit, and the debt‑to‑income ratio sits between 45‑55 %. Traditional credit‑scoring models tend to underestimate risk for this group because they ignore employment volatility and housing instability.

Statistical insights – Interaction analyses reveal that renters are disproportionately dependent on unsecured debt, while owner‑occupiers lean toward mortgage exposure. Multivariate regression shows that a 1 % increase in the share of non‑standard employment raises the average debt‑to‑income ratio by 0.8 % (p < 0.01). These quantitative relationships underscore the importance of incorporating socio‑economic context into risk models.

Practical implications – The authors propose two policy actions. First, targeted debt‑adjustment programs, financial‑literacy workshops, and income‑support initiatives should be directed at the low‑income/high‑debt cluster to pre‑empt defaults. Second, credit‑risk scoring systems should be enriched with variables capturing employment stability and housing tenure, thereby improving predictive power for the special‑debt cluster.

Limitations and future work – The study relies on a cross‑sectional snapshot (pre‑2018) and therefore cannot capture temporal dynamics in household debt behavior. Cluster interpretation also depends on expert judgment. Future research directions include longitudinal panel analysis to observe cluster evolution, the integration of external credit bureau data for validation, and the exploration of deep‑learning‑based automatic labeling to reduce subjectivity.

In sum, the paper demonstrates that a mixed‑type KDD framework can uncover hidden structures in unlabelled consumer‑credit data, produce actionable household risk profiles, and guide more nuanced credit‑risk management strategies.