Abstract

This paper presents the importance of finding the suitable configurations for Artificial Intelligence and Machine Learning algorithms and correct data preprocessing for a waveform problem. In the Artificial Intelligence and Machine Learning area, this step is one of the most important and it influences the performance result of the model. The experiments of different configurations were done using National Instruments Automated Machine Learning (NI AutoML), a web application created for everyone that allows us to easily change the configurations of the model by just clicking some buttons. This work shows how the model performance is influenced by modifying what columns of data to use, by data splitting or by adding or deleting preprocessing steps in the pipeline. All the results obtained for the different experiments are analyzed in this paper. The proposed flow is generic enough to be applied for all the use cases. To exemplify the whole process, a synthetic data set obtained by generating current and voltage in an RL circuit was chosen and the experiments part was created. The data represent two waveforms: one for current and one for voltage and they represent data recorded during the test time. In the end process each test has a label associated: Pass or Fail. The classification problem was defined for help in improving the fail detection rate.

Full paper: click here