Effort estimation is a crucial part of software project planning.  Improper estimates can lead to over budgeting, delays, and project cancellation.  In order to address the need for better software project effort estimates, the field of search based software engineering has been developed.\cite{Harman2001833} Search based software engineering uses data collected from previous projects to make inferences about new projects.
\\
Search based software engineering has grown since its inception, but there are still many open research questions.  Harman suggests that promising areas of future work include hybridizing existing project planning methods, as well as analyzing the terrain of software project data.\cite{Harman2007}  This paper seeks to address both concerns, by employing a combination of algorithms. (COMBA) 
\\
COMBA combines different data preprocessors and learners used in search based software engineering, creating a set of combined algorithms which are a hybridization of different techniques.  COMBA's approach also generates many different algorithms, allowing the data to be assessed from many view points at once.  The behavior of different algorithms on a dataset can give hints as to its appearance and the terrain of the dataset.
\\
Still, the search based software engineering research community does not have a consistent reporting method for the many different algorithms.  In order to address this concern, this paper examines frequently used measures to evaluate algorithms for search based software engineering.  It examines how different error measures rank algorithms, and how algorithms tend to perform across different datasets.  This is to test if a standard is necessary in the literature, or if the results from different error measures are approximately equivalent.  If a certain error measure is more indicative of an algorithms ability to make proper estimates then others, this could have implications on previously published research.
\\
A previous study using COMBA indicated that error measures were approximately equivalent, but that certain datasets were better indicators of algorithm performance then others.  This experiment is performed on a subset of COMBA from the previous study, with fewer algorithms and datasets.  This is to see if the result is applicable on a smaller scale.
\\
In Section 2, the motivation for the experiment and similar projects are discussed. \\
In Section 3, the experimental procedure is detailed, as well as the different algorithms implemented in COMBA, the error measures used, and the datasets used.\\
In Section 4, the results of the experiment are given and compared to the previous study this work was based on.\\
In Section 5, suggestions for future uses of the combination of algorithms approach are given.