NFRs early estimation through software metrics

We propose the use of regression analysis to generate accurate predictive models for physical metrics using design metrics as input. We validate our approach with 40+ implementations of three systems in two development scenarios: system evolution and first design. Results show maximum prediction errors of 1.66% during system evolution. In a first design scenario, the average error is 15% with the maximum error still below 20% for all physical metrics. This approach provides a fast and accurate strategy to boost embedded software productivity and quality, by estimating Non-Functional Requirements (NFRs) during the first design stages.