Forecasting model of rescue drivers using MS-TDSP

The task was to develop a prediction model that optimized the number of rescue drivers being held on standby for the Berlin Red Cross. A baseline model was developed, then a precise prediction model using a few different ML algorithms was proposed in the case study, and the benefits and implementation were discussed further in the discussion section of the paper.
case study
forecast model
ML
statistics
data science
python
Published

May 31, 2023

Executive Summary

The Berlin Red Cross rescue service (BRC) answers to emergency calls when people are in need. On any given day, there are a set amount of rescue drivers residing on standby to answer these calls. When rescue drivers are not able to work due to temporary illnesses, the estimated number of drivers needed in a day can be interchangeable. The BRC allots a flat total of 90 standby-drivers every day; however, the HR planning department struggles with this approach since seasonal weather patterns affect employee health and allocating a flat number of standby- drivers often leads to having not enough or too many drivers standing by. Our firm took up the task of fixing this organizational puzzle. We used our expertise in predictive modeling to develop a solution for the BRC that leverages machine learning (ML) techniques and data science. The solution developed can hereby be utilized by the organization to monitor, refine, and predict their approach to help redistribute budget towards cost efficient business goals.

Challenge

The task is to develop a solution that allows the planning department to assign standby drivers more accurately. Accuracy, in this case, means that there is minimal amount of extra standby drivers assigned that do not get used, while there is a maximized number of days where additional standby drivers do not need to be assigned. In developing this solution, there are many variables that need to be considered. The challenge herein lies in the business’ ability to understand which features of the dataset will help to predict an optimal number of standby drivers in need, which features can properly account for an accurate prediction, and the total volume, value, and variety of the organization’s data.

Solution

Our solution takes the necessary steps to clean and preprocess the accumulation of data in BRC’s data warehouse (DWH) before using it to train ML models that make predictions. As a result of the developments, the company was able to create a solution that not only predicts the optimal quantity of standby drivers needed on a given day, but also improved the coefficient of determination (R2) score from the initial baseline models’ 7.4% to over 99%, while lowering the root mean squared error (RMSE) rates from 33.73 to under 0.01, a quite significant result.

Read the full paper here!

Back to top