Data abstraction plays a crucial role in various application domains, allowing for simplification and representation of complex data sets. This paper focuses on data abstraction in the context of data clustering for test case generation, specifically in the automotive domain. Our main objective is to investigate whether we can use data abstraction to enhance the clustering outcome. We propose different abstraction functions for vehicle sensor data obtained from real-world driving data. We use these abstracted data sets as input to a clustering approach that identifies similar driving scenarios and extracts driving episodes. We evaluate the quality of the clusters using three clustering validation metrics and a Pearson correlation-based metric that assesses the similarity between the extracted driving episodes. To evaluate the effectiveness of data abstraction, we compare the metrics results to those obtained using clustering based on the original data sets comprising numerical data. The findings indicate that data abstraction primarily improves the three clustering validation metrics while delivering nearly comparable results regarding the Pearson correlation-based metric and comes with a substantially reduced runtime.

The testing objective is to find interactions with a system under test leading to unexpected behavior. Such interactions are test cases that can be either manually specified or automatically generated. For the latter, we find many methods and techniques in the research literature, including combinatorial testing or model-based testing. In this paper, we focus on automated test case generation based on models where we are interested in extracting models from available data. In particular, we consider automotive testing, where cars and other vehicles must behave correctly in typical driving situations. The idea is to use available driving data from which we want to extract driving models that we can later use for generating test cases, i.e., arbitrary driving patterns for vehicle testing. Besides outlining the foundations, we discuss the first experimental results we obtain using available open-access driving data.

Testing AI-based systems and especially when they rely on machine learning is considered a challenging task. In this paper, we contribute to this challenge considering testing neural networks utilizing mutation testing. A former paper focused on applying mutation testing to the configuration of neural networks leading to the conclusion that mutation testing can be effectively used. In this paper, we discuss a substantially extended empirical evaluation where we considered different test data and the source code of neural network implementations. In particular, we discuss whether a mutated neural network can be distinguished from the original one after learning, only considering a test evaluation. Unfortunately, this is rarely the case leading to a low mutation score. As a consequence, we see that the testing method, which works well at the configuration level of a neural network, is not sufficient to test neural network libraries requiring substantially more testing effort for assuring quality.