Advancing Operational Intelligence: Predictive Process Monitoring for Imperfect Order Prediction in the Order Life Cycle Management Through Machine Learning and Data Science

Abstract

Abstract: Predictive analytics are vital to modern business processes. Businesses rely on them to analyze, understand, forecast, and make strategic decisions based on future data points. This dissertation focuses on predicting the completion status of ongoing processes within the order life cycle. The order life cycle is a broad series of events that begins with placing an order, all the way through to return or delivery. It’s difficult identifying problems within this lifecycle manually, emphasizing the need for predictive techniques.
By using process mining, we may anticipate the outcome of running instances by adding temporal information to previously recognized process models. We construct configurable process models using time-based sampling from older instances. Our main goal is predictive monitoring tasks - more specifically, being able to predict the end output of an ongoing order in an order life cycle. We use various classification techniques to achieve the goal.
Using actual event logs and incorporating time-based sampling methods gives us a good benchmark for predictions performance against reality. According to our research, the best outcomes are obtained when a bag of activities is included as a feature. We demonstrate how elements in event logs affect classifier selection using qualitative analysis.
The assessment of our prediction using F1 score, and support metrics yields valuable information about the factors that lead to order defects. This information may be used to improve operational intelligence and enable pre-emptive actions throughout the order life cycle.