Modeling Sparse Engine Test Data Using Genetic Programming

We demonstrate the generation of an engine test model using Genetic Programming. In particular, a two-phase modeling process is proposed to handle the high-dimensionality and sparseness natures of the engine test data. The resulting model gives high accuracy prediction on training data. It is also very good in predicting low range data values. However, at least partly due to limitations of the data set, its accuracy on validation data and high range data values is not satisfactory. Moreover, the subject experts could not interpret its real-world meaning. We hope the results of this study can benefit other engine oil modeling applications.