Abstract
Since demand is influenced by a wide variety of causes, it is necessary todecompose the explana-tory variables into different levels, extract theirrelationships effectively, and reflect them in the forecast. In particular,this contextual information can be very useful in demand forecasting with largedemand volatility or intermittent demand patterns. Convolutional neuralnetworks (CNNs) have been successfully used in many fields where importantinformation in data is represented by images. CNNs are powerful because theyaccept samples as images and use adjacent voxel sets to integratemulti-dimensional important information and learn important features. On theother hand, although the demand-forecasting model has been improved, the inputdata is still limited in its tabular form and is not suitable for CNN modeling.In this study, we propose a Tab2vox neural architecture search (NAS) model as amethod to convert a high-dimensional tabular sam-ple into a well-formed 3Dvoxel image and use it in a 3D CNN network. For each image repre-sentation, the3D CNN forecasting model proposed from the Tab2vox framework showed supe-riorperformance, compared to the existing time series and machine learningtechniques using tabular data, and the latest image transformation studies.