More Features vs Simpler Models

Business Context

Engineering teams frequently expand feature sets to improve predictive capability and capture additional business insights. Although richer datasets may increase analytical power, excessive numbers of features also introduce unnecessary complexity, increase computational requirements, complicate maintenance, and reduce model interpretability. Organizations therefore need to identify the information that truly contributes to predictive performance while keeping models efficient and manageable.

AI & Robotics TRIZ Resolution

Instead of continuously adding new variables, organizations should identify high-value features through feature selection, dimensionality reduction, and adaptive feature engineering. Eliminating redundant or low-value information simplifies model architecture while preserving-or even improving-predictive performance and operational efficiency.

Applicable TRIZ Principles

Principle 2 – Taking Out removes redundant or low-value features that add complexity without improving predictions.

Principle 3 – Local Quality prioritizes the most informative features according to their contribution to model performance.

Principle 35 – Parameter Changes continuously refines feature selection criteria as data characteristics and business requirements evolve.

Expected Outcome

Simpler models

Faster training

Lower computational requirements

Easier maintenance

Decision Indicators

Early indicators that excessive features are limiting efficiency include:

Feature engineering requires significant manual effort.

Training time continues increasing.

Model explainability decreases.

Numerous features contribute little predictive value.

Engineering teams struggle to maintain feature pipelines.

Monitoring these indicators supports efficient model design.