Investigation of the COSMIC sizing of real-time embedded and AI software for their usage within a priori and a posteriori contexts for estimation purposes in industry

Abstract

The literature on software estimation and software sizing demonstrates that software development effort has value for organizations. This research aims to operationalize a posteriori measurement of functional size of software with COSMIC-ISO 19761 standard in a company that develops real-time embedded and Artificial Intelligence (AI) software for underground mines. In practice, lack of documentation of the software functional requirements, lack of benchmark data and no usage of standard methods lead to additional constraints and new research challenges. In this research, we explore the issues of a priori measures of software based on partial information, and their uses in estimation models. We use the available resources to identify the functional requirements of developed AI and real-time embedded software solutions and size them with COSMIC method. The size of project for a software release is measured, and an exploratory data analysis (EDA) approach and descriptive statistics are adopted to develop a posteriori estimation models for these projects. In addition, an approximation technique based on iceberg analogy and requirement engineering is designed to assign scaling factor to early functional requirements, using two detailed documented COSMIC case studies. The research results demonstrate that: 1. The size information of the AI and real-time embedded software solutions can be used as one of the independent variables to develop a priori estimation models for future projects; 2. In AI projects, this size information can be used for estimating the amount of effort to develop an AI algorithm in such similar projects; 3. The estimation models proposed based on similarity of characteristics and descriptive statistics can be used to estimate effort in the corresponding groups, based on historical data; 4. The results of functionality-based approximation technique show that the data on past projects can be collected and relevant classification of functionalities can be identified and these scaling ratios can be used as a priori estimation early in the life cycle of a software project.

Date	29 Nov 2020
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Alain Abran (Supervisor)