.TH C4.5 1.SH NAMEA guide to the verbose output of the C4.5 decision tree generator.SH DESCRIPTIONThis document explains the output of the program.I C4.5when it is run with the verbosity level (option.BR v )set to values from 1 to 3..SH TREE BUILDING.B Verbosity level 1To build a decision tree from a set of data items each of which belongsto one of a set of classes,.I C4.5proceeds as follows:.IP "    1." 7If all items belong to the same class, the decisiontree is a leaf which is labelled with this class..IP "    2."Otherwise,.I C4.5attempts to find the best attributeto test in order to divide the data items intosubsets, and then builds a subtree from each subsetby recursively invoking this procedure for each one..HP 0The best attribute to branch on at each stage is selected bydetermining the information gain of a split on each of the attributes.If the selection criterion being used is GAIN (option.BR g ),the bestattribute is that which divides the data items with the highest gainin information, whereas if the GAINRATIO criterion (the default) isbeing used (and the gain is at least the average gain across allattributes), the best attribute is that with the highest ratio ofinformation gain to potential information.For discrete-valued attributes, a branch corresponding to each value ofthe attribute is formed, whereas for continuous-valued attributes, athreshold is found, thus forming two branches.If subset tests are being used (option.BR s ),branches may be formedcorresponding to a subset of values of a discrete attribute being tested.The verbose output shows the number of items from which a tree is beingconstructed, as well as the total weight of these items.  The weightof an item is the probability that the item would reach this point in thetree and will be less than 1.0 for items with an unknown valueof some previously-tested attribute.Shown for the best attribute is:    cut  -  threshold (continuous attributes only)    inf  -  the potential information of a split    gain -  the gain in information of a split    val  -  the gain or the gain/inf (depending on theselection criterion)Also shown is the proportion of items at this point in the treewith an unknown value for that attribute.  Items with an unknown valuefor the attribute being tested are distributed across all valuesin proportion to the relative frequency of these values in theset of items being tested.If no split gives a gain in information, the set of items is madeinto a leaf labelled with the most frequent class of items reachingthis point in the tree, and the message:	no sensible splits  .IR r1 / r2is given, where.I r1is the total weight of items reaching this point in the tree, and.I r2is the weight of these which don't belong to the class of this leaf.If a subtree is found to misclassifyat least as many items as does replacing the subtree with a leaf, thenthe subtree is replaced and the following message given:	Collapse tree for.I nitems to leaf.I cwhere.I cis the class assigned to the leaf..B Verbosity level 2When determining the best attribute to test,also shown are the threshold (continuous attributes only),information gain and potential information for a split oneach of the attributes.If a test on a continuous attribute has no gain or there areinsufficient caseswith known values of the attribute on whichto base a test, appropriate messages are given.(Sufficient here means at least twice MINOBJS, an integerwhich defaults to 2 but can be set with option.BR m.)The average gain across all attributes is also shown.If subset tests on discrete attributes are being used,for each attribute being examined, the combinations ofattribute values that are made (i.e. at each stage, thecombination with highest gain or gain ratio) and thepotential info, gain and gain or gain ratio are shown..B Verbosity level 3When determining the best attribute to test,also shown is the frequency distribution table showingthe total weight of items of each class with:    - each value of the attribute (discrete attributes), or    - values below and above the threshold (contin atts), or    - values in each subset formed so far (subset tests)..SH TREE PRUNING.B Verbosity level 1After the entire decision tree has been constructed,.I C4.5recursivelyexamines each subtree to determine whether replacing it witha leaf or a branch would be beneficial.(Note: the numbers treated below as counts of items actuallyrefer to the total weight of the items mentioned.)Each leaf is shown as:.IR        c ( r1 : r2 /.IR r3 )  with:        \fIc\fR   -  the most frequent class at the leaf        \fIr1\fR  -  the number of items at the leaf        \fIr2\fR  -  misclassifications at the leaf        \fIr3\fR  -  \fIr2\fR adjusted for additional errorsEach test is shown as:.IR        att :[ n1 "%  N=" r4 tree=.IR r5  leaf= r6 +.IR r7  br[ n2 ]= r8 ]  with:        \fIn1\fR  -  percentage of egs at this subtree that are misclassified        \fIr4\fR  -  the number of items in the subtree        \fIr5\fR  -  misclassifications of this subtree        \fIr6\fR  -  misclassifications if this was a leaf        \fIr7\fR  -  adjustment to \fIr6\fR for additional errors        \fIn2\fR  -  number of the largest branch        \fIr8\fR  -  total misclassifications if subtree is replaced by largest branchIf replacing the subtree with a leaf or the largest branchreduces the number of errors, then the subtree is replacedby whichever of these results in the least number of errors..SH THRESHOLD SOFTENING.B Verbosity level 1In softening the thresholds of tests on continuous attributes(option.BR p ),upper and lower bounds for each test are calculated.For each such test, the following are shown:.IP "  *" 4Base errors - the number of items misclassified when the threshold hasits original value.IP "  *"Items - the number of items tested (with a known value for thisattribute).IP "  *"se - the standard deviation of the number of errors.HP 0For each of the different attribute values, shown are:.IP "  *" 4Val <=   - the attribute value .IP "  *"Errors   - the errors with this value as threshold.IP "  *"+Errors  - Errors - Base errors.IP "  *"+Items   - the number of items between this value and the originalthreshold.IP "  *"Ratio    - Ratio of +Errors to +Items.HP 0The lower and upper bounds are then calculated so that thenumber of errors with each as threshold would be one standarddeviation above the base errors..SH SEE ALSOc4.5(1)