Date of Award


Degree Type


Degree Name

Departmental Honors


Electrical and Computer Engineering


Many algorithms have been developed to predict future samples of a signal. These algorithms, such as the recursive least squares predictive filter, rely on the assumption that the system generating the signal can be modeled as a linear system of equations. These systems perform poorly when used to predict signals generated by non-linear systems. To predict a non-linear signal, non-linear methods must be used. Regression trees are a simple form of machine learning that is non-linear in nature and can predict output based on a set of given input. The goal of this capstone project was to develop an algorithm for a regression trees predictive filter capable of predicting a non-linear signa. As this capstone was also an engineering design project it was also the goal to have the algorithm be a part of software system capable of allowing the parameters of the algorithm to be changed for testing. This paper details how the algorithm was developed as well as its results. It was found that using certain non-linear input signals that the regression trees predictive filter performed better at predicting than a traditional linear predictive filter. It was also shown that the regression trees predictive filter was able to adapt to a non-linear signal generated by a changing system. In testing on the changing non-linear signal, the filter was compared to a system which reset its prediction model rather than adapt it like the regression trees predictive filter. The regression trees predictive filter had better performance than this resetting system. This shows that the regression trees predictive filter can adapt to a system in such a way that it learned from it.



Faculty Mentor

Todd Moon

Departmental Honors Advisor

Todd Moon

Capstone Committee Member

Jacob Gunther