In English

Software Prediction Viewer Improving understandability by visualizing future data over time

DUR ABUZAID
Göteborg : Chalmers tekniska högskola, 2017. 95 s.
[Examensarbete på avancerad nivå]

Human brains process and retain information more rapidly when it is presented visually. Information visualization tools have been used by companies to assist an- alytical experts in gaining deeper insights about software data. The present study proposes a design for a visualization tool in the form of a program that uses, pre- dicted/future data to amplify cognition by presenting raw data visually. Compa- nies are currently interested in monitoring software prediction models that provide information regarding software defects levels to enhance understandability of the product status, find reasons for high levels of defects, and thus make informed deci- sions more quickly. This thesis was initiated in light of the interest of the Ericsson Company in Sweden to visualize software prediction model data. The methodology encompasses understanding software prediction models and users’ needs, tasks, and environments. Subsequently, I propose a design that visualizes the requirements regarding what future software data is needed. Finally, an evaluation of the design conducted with the end users to ensure that it fulfills the goal of the thesis, meets users’ needs and in order to make further adjustments. A human- centered design (HCD) approach was followed to facilitate the proposal of the final design of Predic- tion Viewer and meet the goal of this thesis. The design approach was divided into three phases and the result of each phase was used as an input to the next phase, which encourages repeating these phases and frequently refines the solution. The user research involved 16 participants from Ericsson divided into three user groups, where each group represents a user level in the organization’s structure (system, subsystem, and file levels). Of the three focus group sessions conducted with the three user groups, 18 groups of requirements were defined, representing their needs for a system that visualizes future software defects. The most notable reasons are understanding defect causality, software quality, release/assessment management, defects fix planning, test case selections, and highlighting the most error- prone & code smell areas. Given those requirements and needs, a semi-interactive prototype called Prediction Viewer was developed representing information regarding future software defects. Prediction Viewer is a web page that presents information regard- ing defects and defect priority levels over time. Development teams are interested in predicting the number of defects on system and system parts levels; the type of defects; the defect phase found during the testing phases; the answer code assigned to each defect; and the platform where each defect is found. All this information was shown to enhance awareness regarding software status, thus improving the ability to make informed decisions. Finally, a feedback session was conducted with partic- ipants from each user group in order to evaluate the prototype, the importance of the information presented, and the way it was presented. Most of the suggestions were then applied to the final design.

Nyckelord: Software,Visualization,Machinelearning,Futuredata,Softwarepredic- tion, Humancentereddesign,Usercentereddesign.



Publikationen registrerades 2017-08-03. Den ändrades senast 2017-08-03

CPL ID: 250890

Detta är en tjänst från Chalmers bibliotek