J4.8 Example Output

stepweedheightsAI and Robotics

Oct 15, 2013 (3 years and 8 months ago)

86 views



1

WEKA Results

J4.8 Example Output



=== Run information ===
**The keyword list used was generated by evaluating each page under each topic indepentendly using domain knowledge.


Scheme: weka.classifiers.trees.J48
-
C 0.25
-
M 2

-

The classifier run on
test data with options.

Relation: FeatureVector



-

Internal name of the data set.

Instances: 116




-

Total instances in the arff file

Attributes: 80




-

Total number of attributes


Test mode:
evaluate on training data



-

Traning Set, Suppli
ed Test Set, Cross
-
Validation, Percentage Split


=== Classifier model (full training set) ===


J48 pruned tree

------------------


audio <= 0

| agent <= 0

| | machine <= 0

| | | sort <= 0: CompHistory (7.0/1.0)

| | | sort > 0: Search_Sor
t (21.0)

| | machine > 0

| | | card <= 0

| | | | museum <= 0: MachineLearning (24.0/1.0)

| | | | museum > 0: CompHistory (4.0/1.0)

| | | card > 0: CompHistory (8.0)

| agent > 0: Agent (26.0/2.0)

audio > 0

| IBM <= 0: MP
EG (23.0)

| IBM > 0: CompHistory (3.0)


Number of Leaves :

8


Size of the tree :

15



Time taken to build model: 0.05 seconds


=== Evaluation on training set ===

=== Summary ===


Correctly Classified Instances 111 95.6897 %

-

The number

of instances that were correctly classified, the “Score” so to speak. 100% is best.

Incorrectly Classified Instances 5 4.3103 %

-

The number of instance that were incorrectly classified, the failure percentage.

Kappa statistic



0.9461

-

Measures the agreement of prediction with the ture class, 1.0 signifies complete agreement.

Mean absolute error


0.0304

-

The average of difference between predicted and actual values in all test c
ases; the average predicition
error.

Root mean squared error

0.1233


-

Most commonly used as a measure of success of numeric prediction.

Relative absolute error

9.5238 %



-

The total absolute error made rel
ative to what the error would have been if the prediction simply had been
the








average of the actual values.

Root relative squared error

30.8617 %


-

The total squared error made relative to what the error would have been if the pre
diction








had been the average of the absolute value.

Total Number of Instances

116



-

Total instances tested.


=== Detailed Accuracy By Class ===


TP Rate

FP Rate


Precision Recall

F
-
Measure Class


1

0.0
21


0.909


1

0.952


CompHistory


1

0


1


1

1


MPEG


1

0.022


0.923


1

0.96


Agent


0.885

0.011


0.958


0.885

0.92


Mac
hineLearning


0.913


0


1


0.913

0.955


Search_Sort


TP Rate

(
True Positive
) is the proportion of examples which were classified as class “x”, among all samples which are really of class “x”.


Example: The diagonal element

divided by the sum of the relevant row.

FP Rate

(
False Positive
) is the proportion of examples which were classified as class “x”, but belong to a different class among all examples which
are not of class “x”.


Example: The sum of a column in class “x” di
vided by the rows sum of all other classes.

Precision

is the proportion of examples which truly are class “x” among all which were classified as class “x”.

Recall

is equivalent to TP Rate.




=== Confusion Matrix ===


--
The number of correctly class
ified instances is the sum of the diagonals.



a b c d e <
--

classified as


20 0 0 0 0 | a = CompHistory
--
All 20 CompHistory pages were correctly classified into group a (CompHistory)


0 23 0 0 0 | b = MPEG
--
All

23 MPEG pages were correctly classified into group b (MPEG)


0 0 24 0 0 | c = Agent
--
All 24 Agent pages were correctly classified into group c (Agent)


1

0
2

23 0 | d = MachineLearning
--
23 ML pages were correctly classifie
d into group d (Machine Learning), 1 incorrectly in a (CompHistory), 2 incorrectly in c (Agent)


1

0 0
1

21 | e =
Search Sort

--
21
Search Sort

pages were correctly classified into group e (SearchSort),


1 incorrectly in a (Comp
History),


1 incorrectly in d (Machine Learning)

The decision tree constructed by the J48 classifier. This indicates how the classifier uses
the attributes to make a decision. The leaf nodes in
dicate which class an instance will be
assigned to should that node be reached. The numbers in parentheses after the leaf nodes
indicate the number of instances assigned to that node, followed by how many of those
instances are incorrectly classified as a
result. With other classifiers some other output will
be given that indicates how the decisions are made, e.g. a rule set. Note that the tree has
been pruned. An unpruned tree can be produced using the "
-
U" option.