Forecast Modelling via Variations in Binary Image-Encoded Information Exploited by Deep Learning Neural Networks.

Area of Science:

• Artificial Intelligence
• Data Science
• Forecasting Methodologies

Background:

• Traditional forecasting models struggle with diverse data formats (text, voice, image).
• Existing methods often require extensive data analysis and cleansing.
• A need exists for robust forecasting applicable to varied and large datasets.

Purpose of the Study:

• To propose a novel image-encoded forecasting method.
• To leverage deep learning for analyzing image representations of data.
• To demonstrate the model's efficacy on a real-world forecasting task.

Main Methods:

• Decimal data transformed into binary digital two-dimensional (2D) images.
• Raw variables directly converted to input images, omitting data cleansing.
• Convolutional Neural Network (CNN) utilized for pattern recognition in image data.
• CNN employs shared weights, pooling, and back-propagation for nexus identification.

Main Results:

• The image-encoded CNN model successfully processed and analyzed binary digital images.
• The model demonstrated potential for forecasting with vast data volumes.
• Validation on the Global Energy Forecasting Competition 2012 power loads dataset confirmed efficacy.

Conclusions:

• The proposed image-encoded forecasting method offers a viable alternative to traditional models.
• CNNs are effective for identifying patterns in image-encoded data for forecasting.
• This approach shows promise for handling large-scale, multi-format datasets in forecasting.