Software Defect Prediction Based on Fuzzy Weighted Extreme Learning Machine with Relative Density Information
Software defect training set , where ;
Penalty factor , Hidden layer neurons .
Divide into two sets, only contains positive instances, and only contains negative instances. Here, SDP aims to discover the defective modules, so this paper uses positive instances to represent software defective instance, and negative instances to represent non-defective instances;
Count the number of instances in and , then record them as and , where ;
Calculate the class imbalance ratio as ;
Calculate the parameter K for positive and negative class, where ,;
For each instance in , calculate its instances to the th nearest neighbors in and record it as , as well for each instances in , calculate its distance to the th nearest neighbors in and record it as ;
Calculate the relative density of each instance and find the noise in two different classes by equations (10) and (11);
For each instance in , calculate its relative density by equation (8), and then calculate its fuzzy membership value by equation (12);
Obtain based on the value of , and train a FWELM-INTER classifier by equation (5) with the given parameters C and L.