What explains our marks?
Submitted by Anand S (@sanand0) on Wednesday, 24 May 2017
Section: Crisp talk for Data in Government track Technical level: Beginner
The NCERT put together a large-scale survey called the National Achievement Survey. This captured student performance across 4 subjects via 100 questions each, the demographics and behaviour of students, teachers and schools through 300 more questions.
The question was: what affects children’s marks? For example: How does TV watching affect their performance? Is this a bigger effect than playing sports? Is this uniform across states? Do tuitions help or hurt? How does this vary for rich parents vs poor parents?
This talk covers the techniques used to analyse the data, and how this generalises to arbitrary datasets. This has been encapsulated into an open source library called autolysis that we’ll be releasing for the talk.
The intended audience is a beginner to ML who wants to understand how simple algorithms can lead to powerful results if applied the right way. The audience will leave with a specific technique (that we call group-means) that helps identify root causes across any categorical dataset.
Anand is a co-founder of Gramener, a data science company. He leads a team of data enthusiasts with skills in analysis, design, programming and statistics.
He studied at IIT Madras, IIM Bangalore and LBS, and worked at IBM, Infosys, Lehman Brothers and BCG. He and his team explore insights from data and communicate these as visual stories.