Semiconductor Demand Forecasting using Long Short-term Cognitive Networks

Isel Grau, Michiel de Hoop, Ana Glaser, Gonzalo Nápoles, Remco Dijkman

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review


Demand forecasting plays a paramount role in effective supply chain management, giving a business the opportunity to optimize production and improve stock management and operation. Statistical techniques are widely and fittingly used in these prediction problems, however, recent advancements in machine learning techniques are worth exploring. In this paper, we use Long Short-term Cognitive Networks (LSTCN) for forecasting multivariate time-series data describing the demand for six different types of products of a semiconductor company. The results of the experiments show that LSTCN is able to outperform state-of-the-art techniques for three out of the six tested datasets. The results also show that LSTCN is able to leverage the inclusion of additional data and more accurately forecast peaks and valleys in demand. These results and the interpretability potential of LSTCN led to the integration of this algorithm into the suite of forecasting models available in the company’s forecasting system.
Original languageEnglish
Title of host publicationProceedings of the 34th Benelux Conference on Artificial Intelligence and 31st Belgian-Dutch Conference on Machine Learning, BNAIC/BeNeLearn 2022
PublisherAntwerpen University
Number of pages13
Publication statusPublished - 1 Nov 2022


Dive into the research topics of 'Semiconductor Demand Forecasting using Long Short-term Cognitive Networks'. Together they form a unique fingerprint.

Cite this