You have a poorly performing risk model and are looking for strategies to improve performance. You know that only about one percent of your cases represent risk, and you have over 1 million cases to…

The correct answer is D. Use the Ensemble node to reduce the number of non-risky cases used to train the model by a. Option D is correct because when facing severe class imbalance (1% vs 99%) in a large dataset, an Ensemble approach addresses the problem by creating multiple balanced subsets from the non-risky (majority) cases, training separate models on each balanced subset, and combining…

Modeling Techniques

Question

You have a poorly performing risk model and are looking for strategies to improve performance. You know that only about one percent of your cases represent risk, and you have over 1 million cases to use for training purposes. What is the correct approach to test for improving performance?

Options

AUse the Time Series node to reduce the number of all cases by a factor of 100.
BUse the CHAID node to reduce the number of all cases used to train the model by half.
CUse the Balance node to reduce the number of non-risky cases used to train the model by a
DUse the Ensemble node to reduce the number of non-risky cases used to train the model by a

How the community answered

(62 responses)

A
3% (2)
B
15% (9)
C
8% (5)
D
74% (46)

Explanation

Option D is correct because when facing severe class imbalance (1% vs 99%) in a large dataset, an Ensemble approach addresses the problem by creating multiple balanced subsets from the non-risky (majority) cases, training separate models on each balanced subset, and combining their predictions. This leverages the full scale of the 1M+ dataset without permanently discarding majority-class information - a critical advantage with such extreme imbalance.

Option A is wrong because the Time Series node is designed for sequential/temporal data forecasting, not class balancing. Reducing all cases by a factor of 100 preserves the same 1:99 imbalance ratio, so the model still sees the same skewed distribution.

Option B is wrong because CHAID is a decision tree modeling algorithm - it classifies data, it does not sample or balance training sets. It cannot be used as a data preparation strategy.

Option C is wrong because while the Balance node can undersample the majority class, a simple halving still leaves a roughly 1:49 ratio - still severely imbalanced. Ensemble resampling is the more statistically robust tool when imbalance is this extreme and data is abundant.

Memory tip: Think "E for Ensemble = Equal exposure" - with millions of cases and severe imbalance, Ensemble methods give the minority class equal exposure across multiple training rounds, rather than a one-shot undersample that wastes data or still leaves imbalance.

Topics

#class imbalance#Balance node#Ensemble node#model performance

Community Discussion

No community discussion yet for this question.

Full C2090-930 Practice