Iterative process to improve GQM models with metrics thresholds to detect high-risk files

Manual code inspections are intense and time-consuming activities to improve the maintainability and reusability of source code. Although automatic detection of high-risk source code file by metrics thresholds can help inspectors, determining the optimal thresholds is difficult Thus, we propose an iterative process to define and improve GQM models with metrics thresholds to detect high-risk files Our process clarifies experts' viewpoints in the inspection and the measurement metrics using the GQM method, define how to interpret the metrics values, searches concrete thresholds for a specific project by supervised learning using some of the file in the project as training data, and analyzes how to improve models and thresholds. We implemented our tool in R language and evaluated our process using a industrial project. Small-sized embedded C++ systems require only a few training data.