\relax \citation{Frank+Asuncion:2010} \citation{Frank+Asuncion:2010} \citation{Hart67} \citation{Frank+Asuncion:2010} \citation{jalali08} \citation{Frank+Asuncion:2010} \@writefile{toc}{\contentsline {section}{\numberline {1}Introduction}{1}} \newlabel{section:introduction}{{1}{1}} \citation{Devi2002,Bezdek2001,Dasarathy94,Li2009,Bezdek98,Cano2005,Garain2008,Veenman2005,García2008} \citation{Dasarathy94} \citation{Hart68a} \citation{Gates72} \citation{Hart68a} \citation{Gates72} \citation{Hart68a} \@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces Illustration of brittleness in the iris data-set. Before using the instance selector CLIFF (decribed later in this paper) the classes versicolor and virginica show intermingling (brittleness). After using CLIFF, brittleness is reduced.}}{2}} \newlabel{fig:intropic}{{1}{2}} \@writefile{toc}{\contentsline {section}{\numberline {2}Related Work}{2}} \newlabel{section:related}{{2}{2}} \@writefile{toc}{\contentsline {subsection}{\numberline {2.1}Condensed Nearest Neighbor (CNN)}{2}} \@writefile{toc}{\contentsline {subsection}{\numberline {2.2}Reduced Nearest Neighbor (RNN)}{2}} \@writefile{toc}{\contentsline {subsection}{\numberline {2.3}Minimal Consistent Set (MCS)}{2}} \citation{Dasarathy94} \citation{lot2010} \citation{lot2010} \citation{Olvera2010} \citation{Olvera2010} \citation{Olvera2010} \citation{Olvera2010} \citation{FCNN07} \citation{Hart68a} \citation{wilson00} \citation{Ritter75} \citation{Wilson72} \citation{Raicharoen05} \citation{Tomek76} \citation{Kriegel05} \citation{Brighton2002} \citation{Garain2008} \citation{OCM2007,Carrasco05} \citation{Riquelme03} \citation{Raicharoen05} \citation{Narayan06} \citation{Olvera2010} \citation{Olvera2010} \citation{knn} \@writefile{toc}{\contentsline {subsection}{\numberline {2.4}Prototype Selection by Clustering (PSC)}{3}} \@writefile{toc}{\contentsline {subsection}{\numberline {2.5}More IS}{3}} \@writefile{toc}{\contentsline {section}{\numberline {3}Background}{3}} \newlabel{section:bkg}{{3}{3}} \@writefile{toc}{\contentsline {subsection}{\numberline {3.1}Classification with KNN}{3}} \newlabel{subsection:knn}{{3.1}{3}} \@writefile{toc}{\contentsline {subsection}{\numberline {3.2}The Brittleness Measure}{3}} \newlabel{subsection:bm}{{3.2}{3}} \newlabel{eq:bm}{{1}{3}} \@writefile{toc}{\contentsline {subsection}{\numberline {3.3}Noise vs Brittleness}{3}} \citation{jalali08} \citation{Fayyad1992} \citation{jalali08} \citation{jalali08} \@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces Characteristics of some IS (extracted from Table 1 of \cite {Olvera2010}). Also included are some of the published time complexities of the instance selectors.}}{4}} \newlabel{fig:lot-tab}{{2}{4}} \@writefile{toc}{\contentsline {section}{\numberline {4}CLIFF Design and Operation}{4}} \newlabel{section:cliff}{{4}{4}} \@writefile{toc}{\contentsline {subsection}{\numberline {4.1}BORE}{4}} \newlabel{section:bore}{{4.1}{4}} \newlabel{eq:one1}{{2}{4}} \newlabel{eq:one}{{3}{5}} \@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces Pseudo code for BORE}}{5}} \newlabel{fig:sbbra}{{3}{5}} \@writefile{toc}{\contentsline {subsection}{\numberline {4.2}Using Criteria for Instance Selection}{5}} \@writefile{toc}{\contentsline {subsection}{\numberline {4.3}CLIFF: A Simple Example}{5}} \@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces Instance Selection Method used in CLIFF}}{5}} \newlabel{fig:isc}{{4}{5}} \@writefile{lof}{\contentsline {figure}{\numberline {5}{\ignorespaces A log of some golf-playing behavior}}{5}} \newlabel{fig:golf}{{5}{5}} \@writefile{lof}{\contentsline {figure}{\numberline {6}{\ignorespaces Finding the rank of $sunny$}}{5}} \newlabel{fig:rank1}{{6}{5}} \citation{Hart68a} \citation{Dasarathy94} \citation{lot2010} \citation{Frank+Asuncion:2010} \citation{Fayyad1992} \citation{burak} \citation{wilson00} \@writefile{toc}{\contentsline {subsection}{\numberline {4.4}CLIFF: Time Complexity}{6}} \newlabel{section:time}{{4.4}{6}} \@writefile{toc}{\contentsline {section}{\numberline {5}CLIFF Assessment}{6}} \newlabel{section:assess}{{5}{6}} \@writefile{toc}{\contentsline {subsection}{\numberline {5.1}Data Sets}{6}} \@writefile{lof}{\contentsline {figure}{\numberline {7}{\ignorespaces Data Set Characteristics}}{6}} \newlabel{fig:info}{{7}{6}} \@writefile{toc}{\contentsline {subsection}{\numberline {5.2}Experimental Method}{6}} \newlabel{section:brit}{{5.2}{6}} \citation{wilson00} \@writefile{lof}{\contentsline {figure}{\numberline {8}{\ignorespaces Pseudo code for Experiment}}{7}} \newlabel{fig:knnexp1}{{8}{7}} \@writefile{lof}{\contentsline {figure}{\numberline {10}{\ignorespaces Visual illustration of brittleness reduction in the breast cancer data-set for KNN and the IS. Here, the greater the position of each test instance with a predicted target class from an instance with a different target class the better.}}{7}} \newlabel{fig:brit-results}{{10}{7}} \@writefile{toc}{\contentsline {subsection}{\numberline {5.3}Experimental Results}{7}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {5.3.1}Reduction Percent}{7}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {5.3.2}Visualization of Brittleness}{7}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {5.3.3}The Effects of Noise}{7}} \newlabel{}{{5.2}{8}} \newlabel{}{{5.2}{8}} \newlabel{}{{5.2}{8}} \newlabel{}{{5.2}{8}} \newlabel{}{{5.2}{8}} \@writefile{lof}{\contentsline {figure}{\numberline {9}{\ignorespaces Effect of noise on reduction percent. As noise increases from 0 - 40\%, inferior instance selectors generally increase in reduction \%, i.e. the number of instances selected from each training set increases as noise increases.}}{8}} \newlabel{fig:red-results}{{9}{8}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {5.3.4}Is CLIFF significantly better than other IS?}{8}} \@writefile{lof}{\contentsline {figure}{\numberline {12}{\ignorespaces PDS: other - (knn+cliff). Note: {\em negative} values means CLIFF is performing {\em better}.}}{8}} \newlabel{fig:better1}{{12}{8}} \@writefile{toc}{\contentsline {subsection}{\numberline {5.4}Summary}{8}} \@writefile{lof}{\contentsline {figure}{\numberline {13}{\ignorespaces PRECISION: other - (knn+cliff). Note: {\em negative} values means CLIFF is performing {\em better}.}}{8}} \newlabel{fig:better2}{{13}{8}} \@writefile{toc}{\contentsline {section}{\numberline {6}Case Study: Solving the Problem of Brittleness in Forensic Models Using CLIFF}{8}} \newlabel{section:forensics}{{6}{8}} \citation{Walsh94} \@writefile{lof}{\contentsline {figure}{\numberline {11}{\ignorespaces Effects of noise: Performance measures (pd, prec and pf) of CLIFF vs. KNN in the presence of noise.}}{9}} \newlabel{fig:noise}{{11}{9}} \@writefile{lof}{\contentsline {figure}{\numberline {14}{\ignorespaces PFs: other - (knn+cliff). Note: {\em positive} values means CLIFF is performing {\em better}.}}{9}} \newlabel{fig:better3}{{14}{9}} \citation{09NAS} \citation{09NAS} \citation{Karslake09} \citation{Karslake09} \citation{Karslake09} \citation{fastmap} \citation{fastmap} \citation{fastmap} \citation{fastmap} \citation{fastmap} \citation{fastmap} \citation{fastmap} \@writefile{toc}{\contentsline {subsection}{\numberline {6.1}Motivation}{10}} \@writefile{toc}{\contentsline {subsection}{\numberline {6.2}The CLIFF Avoidance Model (CAM)}{10}} \@writefile{lof}{\contentsline {figure}{\numberline {15}{\ignorespaces Proposed procedure for the forensic evaluation of data}}{10}} \newlabel{fig:process}{{15}{10}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {6.2.1}Dimensionality Reduction}{10}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {6.2.2}FastMap}{10}} \@writefile{lof}{\contentsline {figure}{\numberline {16}{\ignorespaces Example of using the cosine law to find the position of $Oi$ in the dimension $k$. Extracted from \cite {fastmap}.}}{11}} \newlabel{fig:fm1}{{16}{11}} \@writefile{lof}{\contentsline {figure}{\numberline {17}{\ignorespaces Projects of points $O_i$ and $O_j$ onto the hyper-plane perpendicular to the line $O_a$$O_b$. Extracted from \cite {fastmap}.}}{11}} \newlabel{fig:fm2}{{17}{11}} \@writefile{toc}{\contentsline {subsection}{\numberline {6.3}Clustering}{11}} \@writefile{toc}{\contentsline {subsection}{\numberline {6.4}Data Set and Experimental Method}{11}} \newlabel{section:brit1}{{6.4}{11}} \@writefile{lof}{\contentsline {figure}{\numberline {18}{\ignorespaces Pseudo code for K-means}}{11}} \newlabel{fig:kmeans}{{18}{11}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {6.4.1}Is CAM a viable forensic model?}{11}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {6.4.2}Does CAM reduce brittleness?}{11}} \@writefile{lof}{\contentsline {figure}{\numberline {20}{\ignorespaces PDS: other - (knn+cliff). Note: {\em negative} values means CLIFF is performing {\em better}.}}{11}} \newlabel{fig:sbetter1}{{20}{11}} \@writefile{toc}{\contentsline {subsection}{\numberline {6.5}Summary}{11}} \citation{09Zadora,09aZadora,06Aitken,04Aitken,02Koons,99Koons} \@writefile{lof}{\contentsline {figure}{\numberline {19}{\ignorespaces Effects of noise: Performance measures (pd, prec and pf) of CLIFF vs. KNN in the presence of noise.}}{12}} \newlabel{fig:snoise}{{19}{12}} \@writefile{lof}{\contentsline {figure}{\numberline {23}{\ignorespaces Position of values in 1NN and CAM population with data set at 3, 5, 10 and 20 clusters.}}{12}} \newlabel{fig:dist3}{{23}{12}} \@writefile{lof}{\contentsline {figure}{\numberline {21}{\ignorespaces PRECISION: other - (knn+cliff). Note: {\em negative} values means CLIFF is performing {\em better}.}}{12}} \newlabel{fig:sbetter2}{{21}{12}} \@writefile{lof}{\contentsline {figure}{\numberline {22}{\ignorespaces PFs: other - (knn+cliff). Note: {\em positive} values means CLIFF is performing {\em better}.}}{12}} \newlabel{fig:sbetter3}{{22}{12}} \citation{09NAS} \citation{me07} \citation{fastmap} \bibstyle{IEEEtran} \bibdata{refs} \bibcite{Frank+Asuncion:2010}{1} \bibcite{Hart67}{2} \bibcite{jalali08}{3} \bibcite{Devi2002}{4} \@writefile{toc}{\contentsline {subsection}{\numberline {6.6}Threats to Validity}{13}} \newlabel{section:ttv}{{6.6}{13}} \@writefile{toc}{\contentsline {section}{\numberline {7}Conclusions and Future Work}{13}} \newlabel{section:conclusion}{{7}{13}} \@writefile{toc}{\contentsline {subsection}{\numberline {7.1}Conclusions}{13}} \@writefile{toc}{\contentsline {subsection}{\numberline {7.2}Future Work}{13}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {7.2.1}Using CLIFF with Other Classifiers}{13}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {7.2.2}Using CLIFF to Optimized Feature Subset Selection}{13}} \@writefile{toc}{\contentsline {subsubsection}{\numberline {7.2.3}Comparing CAM to Other Forensic Models and Forensic data-sets}{13}} \@writefile{toc}{\contentsline {section}{References}{13}} \bibcite{Bezdek2001}{5} \bibcite{Dasarathy94}{6} \bibcite{Li2009}{7} \bibcite{Bezdek98}{8} \bibcite{Cano2005}{9} \bibcite{Garain2008}{10} \bibcite{Veenman2005}{11} \bibcite{García2008}{12} \bibcite{Hart68a}{13} \bibcite{Gates72}{14} \bibcite{lot2010}{15} \bibcite{Olvera2010}{16} \bibcite{FCNN07}{17} \bibcite{wilson00}{18} \bibcite{Ritter75}{19} \bibcite{Wilson72}{20} \bibcite{Raicharoen05}{21} \bibcite{Tomek76}{22} \bibcite{Kriegel05}{23} \bibcite{Brighton2002}{24} \bibcite{OCM2007}{25} \bibcite{Carrasco05}{26} \bibcite{Riquelme03}{27} \bibcite{Narayan06}{28} \bibcite{knn}{29} \bibcite{Fayyad1992}{30} \bibcite{burak}{31} \bibcite{Walsh94}{32} \bibcite{09NAS}{33} \bibcite{Karslake09}{34} \bibcite{fastmap}{35} \bibcite{09Zadora}{36} \bibcite{09aZadora}{37} \bibcite{06Aitken}{38} \bibcite{04Aitken}{39} \bibcite{02Koons}{40} \bibcite{99Koons}{41} \bibcite{me07}{42}