Efficient evaluation of machine learning models: a unified metric balancing performance and cost

This paper introduces a novel, unified metric for evaluating the efficiency of machine learning, deep learning, and artificial intelligence models by balancing predictive performance and execution cost. Existing metrics typically isolate performance or execution measures (e.g., FLOPs, latency, energy), failing to capture the inherent trade-off between resource constraints and predictive capability in single formula. The proposed formula incorporates a tunable trade-off factor and hard constraints on performance and cost, allowing principled comparison across models and deployment settings. Our formulation generalizes prior heuristics and demonstrates clear interpretability, scalability, and hardware awareness.