CSE 300: Software Reliability Engineering Topics covered: Software metrics and software reliability Software complexity and software quality
Introduction
Software quality models Software quality model: Multivariate techniques: Use of software quality models:
Multiple discriminant analysis Classification technique: Derives a linear combination of independent variables that discriminates between the a priori groups such that misclassification error rates are minimized.
Multiple discriminant analysis Introduction: Technique for classifying a set of observations into predefined classes Determine the class of an observation based on a set of predictor or input variables Build a model for a set of observations for which the classes are known. Set known as training set Using the training set, the technique constructs a set of linear functions of predictors such that L = b1x1 + …. + c Where b1, b2, b3,.. Are discriminant coefficients, and c is a constant. Discriminant functions used to predict the membership of an observation with a unknown class. Assign an observation to a discriminant with the highest value.
Multiple discriminant analysis Many techniques: Independent variables are uncorrelated:
Multiple discriminant analysis: Quality model Objects are program modules Independent measures could be measures of complexity Predefined classes based on certain criteria: Complexity metrics could be correlated: Derive principal components, and domain vectors from the metrics. Use domain metrics in the analysis
Multiple discriminant analysis: Quality model Two aspects: Classification errors: Consequences of classification errors:
Multiple discriminant analysis Uncertainty: