AutoUnit:Automatic Test Generation Based on Active Learning and Prediction Guidance

Abstract

Automated test case generation technology aims to reduce test costs.Compared with manual test generation,it has higher test efficiency.Most existing testing tools treat all files in the software equally,but in fact,files with defects account for only a small part of the whole code.Therefore,if testers can detect files that are more prone to defects,they can greatly save testing resources.To solve the above problems,this paper designs a predictive guidance test tool AutoUnit,which is based on active learning.We first predict the defect files in the whole file pool to be detected.Next,we use the detection tool to detect the most “suspicious” files.Then we feed back the actual detection results to the prediction model and update the model to enter the next round of prediction.In addition,when the total number of defective files is unknown,AutoUnit can stop in time by setting diffe-rent target recall rates.It can predict the total number of defective files according to the tested files,calculate the current recall rate,judge whether to stop predict guidance and ensure testing efficiency.Experimental analysis shows that when the same number of defect files are tested,the shortest time and the longest time taken by AutoUnit is 70.9% and 80.7% of the current mainstream testing tools,respectively.When the total number of defective files is unknown and the target recall rate is set to 95%,compared with the latest version of Evosuite,AutoUnit only needs to check 29.7% of the source code files to achieve the same detection level,and its test time is only 34.6% of Evosuite,the cost of testing is greatly reduced.Experimental results show that the method effectively improves the efficiency of test.

Keywords

• test case generation|pool-based active learning|defect predict model|random test|detection efficiency