Summary

Kernel density estimation is a classic statistical technique that is in the same family of techniques as the histogram. It allows the user to extrapolate out from sample data to make insights and predictions about the population of particular objects or events. This extrapolation comes in the form of a probability density function, which is nice because the results read as likelihoods or probabilities. The quality of this model is dependent on two parameters: the bandwidth value and the kernel function. As discussed, the most crucial component of leveraging kernel density estimation successfully is the setting of an optimal bandwidth. Optimal bandwidths are most frequently identified using grid search cross-validation with pseudo-log-likelihood as the scoring metric. What makes kernel density estimation great is both its simplicity and its applicability to so many fields.

It is routine to find kernel density estimation models in criminology, epidemiology, meteorology, and real estate...