A study on improving efficiency and accuracy of software estimation

Abstract

The software estimation framework PROBE in PSP can give very accurate results for TSP. However, PROBE has greater estimating applicability in the design phases of tile software lifecycle than in early phases because it requires a detailed functional breakdown of the software to be estimated. As an alternative, we can select a more economical and available size and effort estimation methods in early phases. This thesis presents the top-down software estimation method that uses function points and COCOMO II model in early phases of typical TSP projects. First, this thesis shows that this method can be as accurate as the PROBE when it comes to estimating size and effort. Secondly, we will illustrate a technique to increase the estimation accuracy. This technique is called ``calibration``. We calibrate COCOMO II model to one academic project in this thesis. Finally, this thesis verifies that when COCOMO II was calibrated using a particular organization``s historical data, it yielded more accurate results than the un-calibrated COCOMO II for that organization. The key result of this research is that COCOMO II with function points is applicable as an early phase estimation method. The accuracy can be continuously improved by calibrating with historical data.