Sild, Sulev, juhendajaMaran, Uko, juhendajaPihor, LisetteTartu Ülikool. Loodus- ja täppisteaduste valdkondTartu Ülikool. Arvutiteaduse instituut2025-10-202025-10-202025https://hdl.handle.net/10062/116874As a part of this bachelor thesis, the application of feature selection methods for evaluating toxicity (pIGC50) of chemicals using artificial neural networks were examined. An overview of the feature selection methods was compiled, and four different methods were analysed while building neural network models. The best results were achieved with a random forest based selection method. The best model had the R2 value of 0.9534 for the training and 0.8128 for the test set. The results were consistent with previous findings and provided a solution for a larger dataset using fewer molecular descriptors than before, therefor creating an opportunity for interpreting the machine learning results.ethttps://creativecommons.org/licenses/by-nc-nd/4.0/kvantitatiivne struktuurianalüüsnärvivõrkakuutne mürgisusTetrahymena pyroformismolekulaartunnusedneural networkacute toxicitymolecular descriptorsbakalaureusetöödinformaatikainfotehnoloogiainformaticsinfotechnologyThesis