University of Twente Student Theses

Login

Forecasting Demand within the Chemical Industry : A comparison of machine learning and traditional forecasting models

Gerhardus, R.J. (2024) Forecasting Demand within the Chemical Industry : A comparison of machine learning and traditional forecasting models.

[img] PDF
2MB
Abstract:Vivochem is a chemical distribution company within the chemical industry. Demand patterns within the chemical industry are subject to irregularities, making forecasting difficult. When forecasts are unreliable, inventory levels increase to maintain the customer service level. The rise in inventory increases inventory costs. This research developed a forecasting model suitable for the sporadic demand patterns. We evaluated five forecasting models: Holt, ARIMA, K nearest neighbour (KNN), light Gradient boosting machine (LightGBM), and neural network (NN) compared to a moving average model as benchmark. We collected 12 years of data on 29% of items (constituting 139 items) responsible for almost 80% of inventory costs. Splitting this data in 80% training and 20% test data. Additionally, we evaluated adding exogenous and feature engineered data. Evaluating model performance with: bias, Mean Squared Error (MSE), Symmetric Mean Absolute Percentage Error (sMAPE), Mean Absolute Error (MAE) and Root Means Squared (RMS). Based on the performances, the best model was a neural network trained on historical demand data. This model reduced inventory by an average of 22,3% of inventory (in weight units) for 65 items and improved the service rate (fill rate) by an average of four percentage points for 19 items.
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:50 technical science in general
Programme:Industrial Engineering and Management MSc (60029)
Link to this item:https://purl.utwente.nl/essays/102627
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page