Multi-Rule Based Ensemble Feature Selection Model for Sarcasm Type Detection in Twitter
Algorithm 4
AIC computation.
Input: Accuracy of various classifiers
Output: Akaike information criterion
(1) For each feature and for every classifier, read the accuracy obtained by each classification algorithm
(2) Compute AIC using the equation
a. 2k-2ln (accuracy) where k = 1 in this case
(3) Select the classifier with the lowest AIC computed for every feature category; let the lowest be AICmin
(4) Compute information loss for the various classifiers using the formula
Exp (AICmin−AIC value)/2
(5) Select the classifier and feature with information loss close to 1 (i.e., the model that can reproduce compared to the given model with very minimal information loss)
(6) Repeat Steps 1 through 5 for combinations of feature categories