This is a TL;DR version. If you'd like to read more about anything mentioned below, head over to the web application itself.
PRX21: QUANTIFIED SELF
"The quantified self (QS) is any individual engaged in the self-tracking of any kind of biological, physical, behavioral, or environmental information."
Definition borrowed from: The Quantified Self: Fundamental Disruption in Big Data Science and Biological Discovery
If you think the title of this scientific article is a little far-fetched, I agree. When I started this project at the beginning of 2021, I wouldn't have guessed that analysis of self-tracked data is actually quite common. My inspiration came from a Reddit post I stumbled upon a few years ago – someone tracked their time for an entire year and analyzed the data to understand themselves better. It seemed like an interesting idea that requires a negligible time investment in exchange for a possibly significant discovery, so I went ahead with it.
TIME
The first major part of what I tracked is time. I divided days into 30-minute intervals and assigned each one of 15 pre-defined time categories, such as sleep, internet, fun, hobbies, and more.
QUANTIFIED SELF (QS)
While time technically falls under QS, I treat the two individually because I came up with 45 variables to track in QS, and grouping them together with time would be messy. These variables are subjectively evaluated attributes of my day and include things like mood, restfulness, productivity, health problems, and others.
With most of the year 2021 behind us, I thought now would be a good time to delve deeper into the data I've collected and hopefully uncover otherwise hidden patterns about how I function on a daily basis. The result is this web application, which is divided into two parts: interactive visualizations and predictive algorithms.
Interactive visualizations
Line charts | Ridgeline plots | Heatmaps | Scatterplots | K-means clustering
Predictive algorithms (also interactive)
Decision tree classifier | CN2 rule induction | Support vector machine | Multiple linear regression | K-nearest neighbors
Acknowledgements
Prakhar Rathi for the multi-page setup in Streamlit
Avik Jain for model and algorithm infographics
This reddit post for the initial inspiration to do this