Workload and human resources forecast with data-driven models

Machine learning is a computer science technique that could be used to simplify tedious data-driven tasks. In particular, Work Force Management is a well-known process that maximizes performance levels and competency for a company, mostly handled by a planner. Despite the technological improvements in the industry, this task hasn’t been tackled by an automated process yet and it remains to be solved by hand with a specialized staff member of the organization. This manual approach can lead to a loss of resources, thereby the goal of this thesis is to find a data-driven approach that solves such task. After gathering the real-world data of a logistic Dutch company and analyzing it with all the processing techniques in place, I was able to generate forecasting models to not only predict the amount of work to be done by the company in question for the upcoming month but also to predict the number of required employees to handle such load. The model was able to improve the existing ones by the organization with a margin of up to 3%, and it showed how the data-driven approach was overall more suitable than the manual mathematical guess by the company. I will be explaining in detail the architecture of both machine learning models, by leading with a step-by-step mathematical background to achieve such results. All the findings will be compared with the ground truth of the company, and various images will be presented to better clarify the retrieved results, explain why dealing with real-life data can lead to unpredictable outcomes, and how I overcame various shortcomings