Regression Multi-Variate Techniques

Regression Multi-Variate Techniques

Multi-variate techniques Have now finished data description and statistical testing will now move to more advanced (multi-variate) techniques: Regression analysis; quantitative description of trends in data - allows for Data analysis and Geostatistics - lecture X interpolation and extrapolation beyond the input data Discriminant function analysis; a means to differentiate groups in a data Discriminant function analysis and clustering set - used to differentiate and classify Principal component and factor analysis; determine directions in a data set to reduce the number of variables and/or look for processes in the data Cluster analysis; group data into homogenous clusters - used to differen- tiate and to split up multi-modal data sets for use in other stat techniques Spatial geostatistics; techniques for mining spatially distributed data Multi-variate techniques: regression Multi-variate techniques Key aspects of regression analysis Have now finished data description and statistical testing will now move to more advanced (multi-variate) techniques: It generates a model of your data; quantitative description of trends in data - allows for interpolation and extrapolation beyond the input data Regression analysis; quantitative description of trends in data - allows for interpolation and extrapolation beyond the input data Strict requirements; normality and no trends or bias in the residuals, no overly influential data points Discriminant function analysis; a means to differentiate groups in a data set - used to differentiate and classify Significant, meaningful and predictive; need to test that the coefficients and model are significant (r ≠ 0, bi ≠ 0), that the equation chose is the most Principal component and factor analysis; determine directions in a data appropriate and that the model is predictive (no overfitting) set to reduce the number of variables and/or look for processes in the data Cluster analysis; group data into homogenous clusters - used to differen- tiate and to split up multi-modal data sets for use in other stat techniques Spatial geostatistics; techniques for mining spatially distributed data Separation and classification of data Separation and classification of data Two main statistical techniques used to separate and classify: The two techniques have a somewhat different focus: Discriminant function analysis - DFA Discriminant function analysis: find a function/vector that best separates the groups in Cluster analysis your data set Goals of these techniques: Cluster analysis: ‣ to separate group samples into clusters based on their similarity majority of statistical techniques cannot be applied to multi-modal data sets: have to split them into homogenous groups. both techniques allow you to quantify the degree of ‣ to classify to what group should a sample be assigned. Examples: soil classi- membership to each cluster fication, rock classification, etc. Use the combination of a variety of characteristics to link unknowns to specific (pre-defined) groups. Discriminant function analysis Discriminant function analysis Examples of discriminant function analysis How do we determine a discriminant function ? 2D case: difference between athletes multi-D case: boundary mapping Need a training set that defines the groups: data with known grouping can directly visualize the DF DF combines multitude of characteris- e.g. a characteristic group of boxers and basketball players tics that are then plotted in space Next: search within this training set boxers boxers granite for the vector that leads to optimal basketball basketball separation players players This function can then be used to classify unknowns schist Discriminant function analysis Discriminant function analysis How do we determine a discriminant function ? a good DF is a function where SSbetween >> SSwithin The vector of maximum separation can be obtained by sum of squares Find the best DF by optimizing the function for maximum SSbetween / SSwithin methodology DF = b0 + b1X1 + b2X2 + b3X3 + b4X4 + .... so let’s have another look at the sum of squares: fitting of the b - coefficients is generally done by iteration and is thus best performed by a computer program. the cumulative deviation from a mean when data strongly correlated: SSwithin the mean not the best descriptor SSwithin: the cumulative deviation of the data from SSwithin SSwithin when calculating the cum. dev. their respective group’s mean - within variance SSbetween SStotal Instead: use the cumulative SStotal SStotal: the cumulative deviation of the data from deviation from the covariance the overall data mean - total variance SSbetween SSbetween trend: the mean vector SSwithin SSbetween: the cumulative deviation of the group SSwithin to work: correlations within groups means from the overall mean - between variance SSwithin have to be similar between groups Discriminant function analysis Discriminant function analysis Not all variables in the DF are necessarily significant requirements for discriminant function analysis: Have to check if each variable adds something to the separating power of the equation - if not: remove the variable from the DF data must be derived from multi-variate normal distributions DF = b0 + b1X1 + b2X2 + b3X3 + b4X4 + ... covariance matrices should be same for each group (the mean vectors should be parallel) How to check for significance: include everything and test the significance using F and tolerance tests, if not: then rerun with subset of significant variables F-tests: does my fit significantly improve by including this variable ? can still apply discriminant function analysis, but the resulting functions will tolerance: is this var’s separation already covered by another var ? not be linear, and significance and goodness-of-fit are much more difficult to assess include variables stepwise and determine how the fit (correct assignment of training set) improves as you add variables Both are affected by the order of inclusion/exclusion of variables ..... Discriminant function analysis Discriminant function analysis DFA to determine the location of a geological boundary Discriminant function analysis with NCSS: check the tutorial the contact between a granite and a schist: tutorial tells you what all input and output two sets for training and a set of unknowns means + requirements 26 major and trace elements have been check for determined on river sediments in this area significance of the variables with F-tests: river sediment compositions are a mixture ‣ removed F-prob of the drainage area, so boundaries are should be < α diffuse ‣ alone F-prob should be < α use these to derive a discriminating check for tolerance function with which to assign the issues with R2 : unknowns and thereby pinpoint the if 1-R2 is low, the var location of the boundary doesn’t add diff Discriminant function analysis Discriminant function analysis all variables are now topographic view significant in the DF no tolerance issues NCSS gives you two discriminant vectors based on these vars score 1 only one sample is assigned incorrectly: outlier score 2 score 1 and 2 are two combinations of the DF is highly significant 5 vars that lead to maximum separation of schist - granite - unknown the groups - two vectors in multi-D space Why separating vectors instead of boundaries ? Discriminant function analysis topographic view score 1 boundaries are rules in the space defined by score 2 the separating vectors score 1 and 2 are two combinations of the 5 vars that lead to maximum separation of schist - granite - unknown the groups - two vectors in multi-D space Discriminant function analysis Cluster analysis Use the vectors to assign the unknowns - not all fit with these groups Group samples into clusters based on similarity Cluster analysis requires substantial user input (selection of number of clusters, clustering routine, similarity criteria, etc) and results can therefore be ambiguous: always give detailed information on how your cluster analysis was performed Cluster analysis Cluster analysis - sample assignment criteria Group samples into clusters based on similarity range of techniques that can be used to determine similarity whichever deviation between sample cluster and cluster mean is smallest: Wide range of techniques - see book for details mean A cluster assigned to that cluster mean A ‣ Euclidian distance - r or r2 ‣ city block of Manhattan distance - this is _ Cluster analysis is again controlled by cluster (xi-xA) useful when the two variables are separate mean B the sum of squares: cluster _ characteristics (fossil length and width, the (xi-xB) mean B SSwithin SSbetween A-B diagonal is not of interest) SSwithin SSbetween B-C ‣ correlation similarity - sample with the _ (xi-xC) same correlation are grouped together: cluster SSwithin SSbetween A-C deals with dilution effects cluster mean C small variance within: tight clusters mean C ‣ association values - especially useful large variance between: good separation when you have only presence/absence data - specialized increasing the number of clusters will decrease the within variance, until all samples are their own cluster. That result is however meaningless.... Cluster analysis - two types Hierarchical cluster analysis Two varieties of clustering: hierarchical and partitioning methods An example of hierarchical clustering: the composition of a number of lava samples from Kawah Ijen volcano: hierarchical techniques: represent similarity in a tree or dendrogram the method: degree of dissimilarity sample dissimilarity based upon 1. all samples are a separate cluster KV01 nearest

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    10 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us