Prediction of click through rates using hybrid kalman filter-tree structured markov model classifiers

A hybrid prediction computer system for processing training data to predict click-through-rates, said hybrid prediction computer system processing training data by:

• creating a machine learning based model for making a base prediction, said machine learning model constructed using a first set of features in training data;
• creating a tree-structured statistical table from a second set of features in training data by applying Kalman-filter methods to a tree-structured Markov model to estimate parameters in said tree-structured statistical table;
• executing said machine learning model using said first set of features from test data;
• and
• adding an adjustment factor from said tree-structured statistical table to augment said base prediction from said machine learning based model to predict click-through-rates.

A computer-readable medium, said computer-readable comprising a set of computer instructions for implementing a hybrid method of predicting click-through rates, said computer instructions performing the steps of:

• creating a machine learning based model, said machine learning model constructed using training data having a first set of features;
• creating a tree-structured statistical table from a second set of features in training data by applying Kalman-filter methods to a tree-structured Markov model to estimate parameters in said tree-structured statistical table;
• executing said machine learning model using said first set of features from test data;
• and
• adding to augment the base prediction of the executed machine learning model by an adjustment factor obtained from said tree-structured statistical table using said second set of features from said test data to predict click-through-rates.