Training is a powerful and required method to equip human resources with tools to keep their organizations competitive in the markets. Typically at the end of class, trainees are asked to give their feelings about or satisfaction with the training. Although there are various reasons for conducting training evaluations, the common theme is the need to continuously improve a training program in the future. Among training evaluation methods, post-training surveys or questionnaires are the most commonly used way to get trainees’ reaction about the training program and “the forms will tell you to what extent you’ve been successful.” (Kirkpatrick 2006) A higher satisfaction score means more trainees were satisfied with the training. A total of 40 prediction models grouped into 10-GIQs prediction models and 6-GIQs prediction models were built in this work to predict the total training satisfaction based on trainees’ general information which included a trainee’s desire to take training, a trainee’s attitude in training class and other information related to the trainee’s work environment and other characteristics. The best models selected from 10-GIQs and 6-GIQs prediction models performed the prediction work with the prediction quality of PRED (0.15) >= 99% and PRED (0.15) >= 98%, separately. An interesting observation discovered in this work is that the training satisfaction could be predicted based on trainees information that was not related to any training experience at all. The dominant factors on training satisfaction were the trainee’s attitude in training class and the trainee’s desire to take the training which was found in 10-GIQs prediction models and 6-GIQs prediction models, separately. / text

Целью данной дипломной работы является исследование факторов, влияющих на качество предсказания ретросинтетических схем с помощью инструмента AiZynthFinder. В ходе тестирования модели были подобраны оптимальные параметры запуска анализа, позволившие увеличить долю решенных молекул с 40% (для базовых условий) до 80% (для оптимальных условий). Инструмент на основе AiZynthFinder был интегрирован в собственный вычислительный веб-сервис. / The goal of this work is to study the factors affecting the quality of retrosynthetic schemes prediction using the AiZynthFinder tool. During the model testing, optimal parameters for running the analysis were found, which allowed increasing the fraction of solved molecules from 40% (for baseline conditions) to 80% (for optimal conditions). The AiZynthFinder-based tool was integrated into an in-house computational web service.