Noise-injection as an Approach to Generating Random Data Sets for Online Tests and Virtual Labs

Abstract

A methodology based on noise injection for generation of randomized tabulated data is presented. The strategy can be used both in online teaching and for specific numerical/graphical exercises, when individualized data sets are required simultaneously for the students. Restrictions imposed on teaching methods during to the SARS-Covid-19 pandemic, especially for laboratory sessions in chemistry or even for preparing written exams, have led to a need for approaches based on randomized data sets based on literature data or theoretical equations. Commonly available spreadsheet software has been used for generating random data and for analysis and calculations, which facilitates easy and low cost application of the methodology presented here. Uniform and Gaussian distributions have been employed to generate different types of noise. Statistical analyses on linear regression parameters for the different distribution and levels of injected noise have been performed. As examples, these results are employed to introduce randomness in three typical experiments performed in Physical Chemistry labs involving thermodynamics, chemical kinetics and conductivity of electrolyte solutions. Literature values are employed for the experiments as templates to which different levels of noise are applied. The results indicate that the application of noise has to be carefully controlled. Uniform noise is suggested for data sets that already contain natural random noise, whereas Gaussian noise should be employed for data sets created directly from theoretical or empirical equations, so as to produce data sets with a more natural, realistic appearance.