The purpose of the thesis is a (semi)automatic detection of rapid changes in the cell membrane capacitance signal, namely changes that are a consequence of exocytosis and endocytosis.
Corresponding changes in the signal were previously detected automatically, using mathematical signal analysis. However, mathematical analysis is only possible when the signal-to-noise ratio is sufficiently large.
In some cell types the amplitude of the change is so small that it is almost completely hidden in the noise, therefore simple mathematical analysis is no longer possible.
Consequently, the detection of the corresponding rapid changes is currently done manually. A number of signals have been labeled and so offers a possibility of facilitating the manual inspection by applying machine learning, which is the work of this thesis.
The data is initially pre-processed, where calibration pulses and areas with extensive noise are removed from the measurements and the baseline of the signal representing the measurement is corrected.
This is followed by a preparation of the dataset, which includes the extraction of instances, calculation of the attributes of instances and resampling of the training dataset. Then, based on the training set, test set and a variety of machine learning methods, results of the classifications of new cases are presented and compared.