Tazama rule results include an "indpdntVarbl" field alongside the weight and sub-rule reference. What does the Independent Variable represent, and where is it documented?
1 Answers
The Independent Variable (indpdntVarbl) is the raw, machine-readable result of a rule processor, delivered alongside the human-readable result band.
Rule processors typically deliver two results:
- A result band - the human-readable interpretation of the behaviour the rule evaluates (good, neutral, bad).
- An Independent Variable - the raw value the rule measured. If a rule counts the number of transactions a creditor received in 24 hours, the Independent Variable is that count; if a rule determines how long it has been since the debtor last performed a transaction, the Independent Variable is the elapsed time in milliseconds.
The two are directly related: the rule processor uses the Independent Variable to look up which band the result falls into. The Independent Variable is the machine-readable interpretation of the behaviour; the band is the human-readable one.
The field exists to enable AI/ML modelling of typologies based on rule processor results, using methods such as Linear Regression, Decision Trees, and Random Forest. Statistical models predict the value of a dependent variable from the values of independent variables in the model formula, and model training determines the optimal coefficients for each independent variable to predict a specific outcome (see Linear regression). Exposing the raw rule result as an Independent Variable gives those models the untransformed signal to train on, rather than only the banded interpretation.
The Independent Variable was added to all rule processors as a platform-wide enhancement; there is currently no dedicated public documentation page for the field beyond this explanation.