Data Science interview questions and answers

Poisson models counts of events in a fixed interval when events are rare, independent, and have a constant average rate.

Binomial models the number of successes in a fixed number of independent Bernoulli trials with constant probability.

Products of many positive random shocks, or processes with multiplicative growth, tend toward lognormal.

Kurtosis describes tail heaviness and peak relative to a normal curve, with excess kurtosis often reported so normal equals zero.

A stationary series has a stable mean, variance, and autocovariance structure over time.

Lag features are past values of the series or of covariates, such as yesterday's demand or last week's price.

A rolling mean averages the last k points to smooth noise and reveal slower structure.

Random splits leak future information into training and overstate accuracy.

Retention is the share of a starting cohort that is still active after a stated window, such as day-7 return.

DAU and MAU count unique active users in a day or calendar month under a defined activity event.

A DataFrame is a labeled table with rows and columns, like Excel in code.

I first ask why data is missing.

Histograms show distribution of a numeric variable in bins.

It shows median, quartiles, and outliers quickly.

Correlation is association

Running many tests until one looks significant.

GROUP BY email or phone HAVING COUNT(*) > 1, then investigate.

Comparing magnitudes across categories.

Drop when rare and random