A researcher at Microsoft Research has put developers under 15 EEG, eye-tracker and electrodermal detector. Loaded with this battery of sensors, it invited them to analyze some programming problems and observed their reactions. The objective of the study is to identify times when the developer faced a difficulty and therefore is likely to make mistakes. Objective: Debugger code at a time when the developer is writing it.
How to assess the level of stress of the developer
“And how Does that make you fell” was the title of the presentation by Andrew Begel, senior researcher at Microsoft Research, Erin Solovey, attending Drexel University and Mary Czerwinski, senior researcher at Microsoft Research Professor at the Faculty Summit held last July at the Microsoft campus. This is the annual conference or Microsoft researchers present their joint research with the academic sector. All three presented their research related to the measurement of stress in different areas. Andrew Begel was interested in the stress of the developers. Its purpose is to study the possible correlation between their emotional state and their perception of the difficulty of a task, possibly predict whether a code will be marred by bugs because the developer saw its stress increase before a section of code.
The researcher selected 15 C # programmers (including one woman) with 8 computer to solve problems. Our developers have been fitted with an EEG Neurosky Mindband, a sensor and electrodermal an eye-tracker placed in front of them after they look at the screen, but also measured the size of their pupils. Theoretically, when a difficulty meeting our pupil dilates a few tenths of millimeters. The electrodermal sensor (or EDA) detects sweat. It appears in the effort, but also the degree of stress.
Meanwhile, developers have responded to the questionnaire NASA TLX, the aerospace sector standard questionnaire to measure the workload of a pilot, for example. The researcher then developed three algorithms of Machine Learning to analyze their data, working on information produced by one, two or three sensors or various combinations. On average, the accuracy of predictive algorithms was around 65%, data from the eye-tracker is the most valuable, with the most convincing pair when the eye-tracker in the EDA sensor results.
Stress levels actually correlate with difficulty writing the code
The study showed that Andew Begel analysis of sensor data were correlated with the NASA TLX questionnaire information. One can actually estimate the difficulty of the work delivered via sensors.
The researcher also experimenting with a keyboard that analyzes the user’s typing or mouse which analyzes the hand on her. Most people support with their palm on the mouse when stressed. This may be using all these collector arrays that Visual Studio will report one day that the developer is about to insert a bug in its code. The purpose of the study is threefold. If we can detect that a developer is in a difficult moment in the writing of the code, then perhaps we might prevent interruption of his work currently accurate. Maybe it could be the signal to refactor the code part that brings down the productivity of developers. Translation : Google Translate Source : Présentation « And how does that make you feel? », Faculty Summit 2014 On Demand , July 15, 2014