The Robustness of Machine Learning Models Using MLSecOps: A Case Study On Delivery Service Forecasting

Forecasting delivery services is a key aspect of modern delivery service operations that significantly contributes to the optimization of operations and the enhancement of customer satisfaction. Machine learning can assist in predicting the delivery time. One method to enhance security in machine learning is the implementation of MLSecOps. MLSecOps, or Machine Learning Security Operations, streamlines the process of deploying, monitoring, and maintaining machine learning models to ensure consistent and reliable performance in production environments. Cybersecurity was also integrated to enhance the security, robustness, and resilience of these models. This study applies MLSecOps to forecast delivery services to enhance the robustness of machine learning models. The MLSecOps tool utilized is the Adversarial Robustness Toolbox (ART). The results of testing the machine learning model on Forecasting Delivery Services show robustness to attacks such as boundary and backdoor attacks.

I. Introduction

The rise of e-commerce and shifting consumer behavior have led to a surge in demand for delivery services [1]. This trend has generated a wealth of data that holds valuable information about patterns and trends in service demand. Utilizing this data effectively has become crucial for companies to forecast and anticipate future delivery requirements. Known as Delivery Service Forecasting [2], this process plays a fundamental role in optimizing and managing modern delivery service operations.