What it was
A short structured exercise from September 2024: predict which students are at risk of dropping out, and make the analysis usable by someone who does not read notebooks. The first week was data wrangling, handling missing values, outliers, types, encoding and derived features, then descriptive statistics, correlation analysis and hypothesis tests on questions such as whether admission grades or financial aid relate to dropout. The second week was exploration and visualisation, from univariate plots to principal component analysis, and a Streamlit dashboard that summarises the insights and lets a user explore the data and the risk predictions. The app runs on Streamlit Community Cloud and I wrote the project up on Medium.
What I learned
Hypothesis tests kept me honest: several patterns that looked obvious in a chart did not survive a t-test or a chi-square test, so I learned to lead with the question and let the plots serve it. Building the dashboard taught me that the last mile, deployment and a clear interface, is what turns an analysis into something an administrator would act on. I also started writing about my work in public, which I have kept doing since.