actionable analytics why, how?
TRANSCRIPT
![Page 1: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/1.jpg)
Actionable Analytics: why? how?
TIM MENZIES, YE YANG, [email protected]@gmail.com
LINCOLN, NEBRASKA, Nov, 2015http://action15.github.io
1
![Page 2: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/2.jpg)
How we got here
2001
● “Stop telling me what is. Tell me what to do.”
● -- Gruff user, NASA, dismissing a decision tree
2007
● Norman Fenton, PROMISE keynote
● “Most metrics irrelevant to the industrial mix.”
2012
● ICSE Goldfish bowl panel on predictive analytics.
● “Enough mere prediction. Give us something we can use.”
2
![Page 3: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/3.jpg)
The collapse of prediction?
BEFORE
● Plan-driven process-based estimating approaches
● Requirements => analysis (sizing, effort, schedule) => number (2.71 years)
● E.g. waterfall model, Spiral, RUP, etc.
NOW
● Feature-based development
● Recognition / exploring/ reaction to opportunity
● No longer:
● “We will craft this diamond with these capabilities.”
● Rather:
● “We will explore N possible features, and will deliver M < N”
● “Every feature is an experiment.”
3
![Page 4: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/4.jpg)
End of the era of“the” prediction
● Rather:● PredictionS● Plural
● Range of options ● assessed via criteria ● learned interactively with
business users
4
![Page 5: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/5.jpg)
Exact predictions are spurious?
● Less numbers, more insight● Burak Turhan’s “The graph”● circle = reported to● red = error report● green = error fix● blue = report+fix in the same team
● More coarse grain control
● (“ontime”, “aLittleLate”, “wayOverdue”)● E.g.. Predicting delays in software projects using
networked classification● Choetkiertikul et al. ASE’15 (Thurs morning)
5
![Page 6: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/6.jpg)
6
![Page 7: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/7.jpg)
“What’s that?”(anomaly detection)
● E,g, Keogh’s SAXrepresentation
● Monitors (e.g.) thousands of on-board rocket sensors
7
![Page 8: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/8.jpg)
8
![Page 9: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/9.jpg)
“What if”:e.g. Bayes nets
● All links bi-directional
● Tickle anything to see impact on anything else
● E.g. Fenton et al., TSE, 2000, Misirli et al, TSE 2014
9
![Page 10: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/10.jpg)
“What if”e.g. Pareto clustering
● Range of options
● assessed via criteria
● learned interactively with business users
10
![Page 11: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/11.jpg)
11
![Page 12: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/12.jpg)
“What to do” (e.g. contrast set learning)
● Not predictors for separate classes
● But deltas between classes (much shorter)
● Minwal, STUCCO, TAR2, TAR3
12
![Page 13: Actionable analytics why, how?](https://reader034.vdocuments.net/reader034/viewer/2022042723/587d35bf1a28ab2a448b5e2f/html5/thumbnails/13.jpg)
13