Improving search based test data generation using multi directional search

Abstract

Automated test data generation has been a major academic interest. We present Multi Directional Search (MDS), a new search-based test data generation algorithm. Unlike, Alternating Variable Method, the most commonly used search-based test data generation algorithm, traverses the variable in the function's input vector and change the value of the one variable at a time, MDS changes the value of multiple variables simultaneously. We implemented MDS on CAVM, a search-based test data generation tool for C which uses Extended AVM to handle dynamic data structure. MDS is developed to attached to and used prior to Extended AVM. We call MDS attached Extended AVM "Extended AVM+". We compare Extended AVM+ against Extended AVM using 15 functions and 55 test targets. As a result, Extended AVM+ was able to generate test data with less costs than Extended AVM.