EP005
Adopt the simplest model meeting accuracy thresholds, documenting explainability evidence required by the EU AI Act's transparency obligations.
CyberTRIZ analysis · Process contradiction EP005 · one of 8,235 worked contradictions published by CyberTRIZ.AI
Regulations
Higher Predictive Analytics Accuracy vs. Lower Model Complexity
Business Context. More accurate predictive models forecasting process delays or non-compliance typically require greater complexity, incorporating more variables and more sophisticated algorithms, but complex models are harder to maintain, explain, and retrain reliably.
Process TRIZ Resolution. Rather than maximizing model complexity to chase incremental accuracy gains, organizations should identify the smallest set of variables that captures most of the predictive value, accepting a modest accuracy trade-off in exchange for a simpler, more maintainable model.
Applicable TRIZ Principles
Principle 2 (Taking Out) removes low-impact variables that contribute little to predictive accuracy.
Principle 3 (Local Quality) applies additional model complexity only where it delivers proportionate value.
Principle 35 (Parameter Changes) balances model complexity against maintainability requirements.
Expected Outcome
Sufficient predictive accuracy
Lower model complexity
Easier maintenance and retraining
More explainable predictions
Decision Indicators
Predictive models have grown so complex that no one fully understands their logic.
Model maintenance consumes more effort than the improvement in accuracy justifies.
Retraining the model has become an increasingly difficult, specialized task.
Explaining model predictions to stakeholders has become impractical.
Incremental complexity additions have produced diminishing accuracy returns.
If several of these indicators are present, the contradiction is likely active and the Process TRIZ resolution above should be evaluated.