Insurance Company: Customer Profiles for Caravan Insurance Prediction

TICDATA2000 is a dataset supplied by Sentient Machine Research based on a real-world business problem. It contains 5,822 customer records for training, each with 86 variables including socio-demographic data derived from zip codes and product ownership information. The target variable is 'CARAVAN: Number of mobile home policies', with a separate evaluation set of 4,000 records.

Use Cases

• Predicting caravan insurance policy ownership based on customer socio-demographic and product usage data.
• Analyzing the relationship between zip-code-derived socio-demographic variables and insurance product ownership.
• Building and validating binary classification models for a real-world business problem in the insurance domain.
• Studying feature importance for a specific insurance product within a broader portfolio of customer holdings.

Strengths

• Based on a real-world business problem from a data mining company, suggesting practical relevance.
• Contains 5,822 training records and a separate 4,000-record evaluation set for model validation.
• Includes 86 variables per record, combining socio-demographic and product ownership data.
• Has no missing values, which may simplify preprocessing.

Limitations

• Column-level documentation is absent from this metadata; field semantics must be inferred from an external dictionary.
• Row count for the full collection is not explicitly stated, requiring summation of the split files.
• Socio-demographic data is aggregated at the zip-code level, which may limit individual-level insights.

Provenance

Source

Sentient Machine Research, supplied via UCI KDD Archive.

Collection Method

Derived from real customer data, with socio-demographics inferred from zip area codes.

Time Range

Likely circa 2000, based on the dataset naming convention.

Freshness

Last updated date is unknown; freshness unverified.

Geography

Presumably the Netherlands, given the source company, but not explicitly confirmed.