Predikce času dokončení výdeje objednávek s využitím strojového učení

This thesis deals with predicting the completion of warehouse order picking based on historical operational data. The main objective is to develop and test predictive models based on machine learning methods that can predict the completion times of individual order dispatches using historical data on warehouse staff performance and material movements. The proposed solution follows the CRISP-DM methodology and involves the cleaning, transformation, and merging of three real-world datasets, including the creation of new derived variables and the segmentation of orders using the K-means algorithm. All data analytics and experiments were conducted in the Python programming language using the interactive Jupyter Notebook environment. Subsequently, classification and regression machine learning algorithms were tested, with the Random Forest model emerging as the best performer. During a six-fold cross-validation, the regression model achieved a mean absolute error (MAE) of 3.06 hours (with R² = 0.78). In the classification approach, the model predicted the assignment of orders into time windows with an accuracy of 75.7 %. The comparison revealed that the optimized regression approach is more suitable for operational warehouse management and carrier coordination.