\relax 
\@writefile{toc}{\contentsline {section}{\numberline {1}Results}{1}}
\newlabel{sec:results}{{1}{1}}
\@writefile{toc}{\contentsline {subsection}{\numberline {1.1}Evaluation of AR, MdMRE and Pred(25) Results}{1}}
\newlabel{subsec:mre-based-results}{{1.1}{1}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {1.1.1}Results for Nasa93}{1}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {1.1.2}Results for Desharnais}{1}}
\@writefile{toc}{\contentsline {subsection}{\numberline {1.2}Evaluation of WIN-TIE-LOSS Results}{1}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {1.2.1}Results for Nasa93}{1}}
\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces MdMRE and Pred(25) results for Nasa93 dataset. Neither change of kernel nor the change of bandwidth generates a considerable difference in results. Furthermore, small changes in MdMRE and Pred(25) values due to different kernel-bandwidth combinations do not follow a certain pattern. Another conclusion from this figure is that N-ABE methods fail to improve ABE0.}}{2}}
\newlabel{fig:mre-nasa93}{{1}{2}}
\bibstyle{abbrv}
\bibdata{myref}
\bibcite{Alpaydin2004}{1}
\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces MdMRE and Pred(25) results for Desharnais dataset. None of the different kernel-bandwidth combinations can improve the performance of N-ABE methods to a point where they outperform the U-ABE method of ABE0.}}{3}}
\newlabel{fig:mre-desharnais}{{2}{3}}
\@writefile{toc}{\contentsline {subsubsection}{\numberline {1.2.2}Results for Desharnais}{3}}
\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces Win-tie-loss results for Nasa93. Results we have for Nasa93 are very similar to Cocomo81 dataset: Neither changing kernels nor the bandwidths provides a noticable change in win-tie-loss values. Also uniform weighting always outperform non-uniform weighting.}}{4}}
\newlabel{fig:win-tie-loss-nasa93}{{3}{4}}
\@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces Win-tie-loss results for Desharnais. The implications we have observed in Cocomo81 and Nasa93 repeat for Desharnais dataset: Change of kernels does not provide a significant change in win-tie-loss values and neither does the change of bandwidths. There are only some small changes in different kernel-bandwidth combinations but we can not observe a pattern. Furthermore, ABE0 has a better estimation performance than all N-ABE methods.}}{5}}
\newlabel{fig:win-tie-loss-desharnais}{{4}{5}}