In a data model, the level of predictive power of an attribute is often measured by information gain. Which option correctly identifies Information gain?

• A. Lift
• B. Information gain
• C. Statistical relevance
• D. Precision factor

Answer: (B)

Information gain measures how much knowing an attribute reduces uncertainty about the target variable. It’s calculated as the dataset’s entropy before the split minus the weighted sum of entropies of the subsets after splitting by that attribute (IG = H(T) − sum_v (|T_v|/|T|) H(T_v)). When an attribute cleanly separates classes, the resulting subsets are purer, and the information gain is high; if the attribute adds no new information, the gain is zero. This is the standard way to quantify an attribute’s predictive power in decision-tree learning. Other terms like lift assess association strength between variables rather than the target’s predictiveness, and precision factor or statistical relevance aren’t the established measure for this purpose. So information gain is the correct concept.