A company in the medical services industry faces challenges in efficiently assigning technicians to service orders, given the variety of competence requirements, availability constraints, and geographic considerations. Manual planning is time-consuming and inconsistent, motivating the design of a machine learning-assisted planning tool. The research was guided by a main question on how machine learning could support planning, design questions on preprocessing, modelling, and system integration, and knowledge questions on class imbalance handling and evaluation metrics. The system was evaluated on functional, machine learning, and technical success criteria. Core achievements include sub-30 second inference times, ERP integration, interpretable recommendations, F1-scores above 0.75, and containerized deployment readiness. The project followed a Design Science methodology, ensuring both rigor and relevance, while CRISP-DM structured the data-driven stages: business understanding, data preparation, modelling, evaluation, and deployment design. Exploratory Data Analysis revealed missing data and inconsistencies, which were addressed through preprocessing. Among the tested models, Random Forest with max depth 3 was selected as the final model due to its robust performance (F1 = 0.883, Accuracy = 0.885) and interpretability. Feature importance analysis showed that competence and availability were the most decisive factors, aligning with stakeholder expectations.

This document contains a master's thesis as authored by a student at Saxion University of Applied Sciences. The grade received by the student is not published here. The complexity or quality of research required of students may vary by programme, as does the minimum required study period (in this case 20 EC). The full text may or may not be publicly available, depending on the agreements with the organization hosting the student for this assignment. Copyright and moral rights for this document are retained by the author and/or other copyright owners.