Dynamic Test Case Selection using Machine Learning

Detta är en M1-uppsats från KTH/Hälsoinformatik och logistik

Författare: Niklas Höstklint; Jesper Larsson; [2021]

Nyckelord: ;

Sammanfattning: Testing code is a vital part at any software producing company, to ensure no faulty code slips through that could have detrimental consequences.  At Ericsson, testing code before publishing is a very costly process which can take several hours. Currently, every single test is run for all submitted code.  This report aims to address the issue by building a machine learning model that determines which tests need to be run, so that unnecessary tests are left out, saving time and resources. It is however important to find the failures, as having certain failures pass through into production could have all types of economic, environmental and social consequences. The result shows that there is great potential in several different types of models. A Linear Regression model found 92% of all failures within running 25% of all test categories. The linear model however plateaus before finding the final failures. If finding 100% of failures is essential, a Support Vector Regression model proved the most efficient as it was the only model to find 100% of failures within 90% of test categories being run. 

  HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)