Four recurring research reflexes
Compact notes connected to working case studies. Each one states the object, the practical consequence and the checks required before trusting a result.
01
Credit risk modelling
A score is useful only inside a declared evaluation protocol. The case study separates the training population, validation split, class balance and threshold decision before comparing models.
Before trusting the result
02
Customer segmentation
K-Means is a compact objective, but scaling and feature geometry determine what a centroid means. Cluster stability and business interpretability matter more than a decorative two-dimensional projection.
Before trusting the result
03
Financial time series
Price levels may be close to unpredictable while volatility remains persistent. GARCH models conditional variance and makes the risk object explicit instead of claiming directional foresight.
Before trusting the result
04
Monte Carlo quantiles
Importance sampling changes the proposal distribution to visit rare but consequential regions more often, then reweights observations to preserve the target expectation.
Before trusting the result