Utilizing Deep Learning to Influence Design Decisions and Predict Future Scenarios

Abstract

Deep learning is increasingly transforming design practice by enabling data-driven decision-making, predictive analysis, and the automation of visually complex tasks. This study investigates the application of a Convolutional Neural Network (CNN) to the CIFAR-10 dataset to demonstrate how image-based deep learning models can support design analysis and future-scenario prediction across fields such as architecture, product development, and urban planning. The model was developed using a sequential CNN architecture with convolutional, pooling, batch normalization, and dropout layers and trained over 20 epochs using the Adam optimizer. Performance evaluation employed accuracy, precision, recall, F1-scores, confusion matrices, and ROC–AUC curves to provide a transparent and interpretable assessment of model behavior. The CNN had a training accuracy of 89% and a test accuracy of 77%. Its macro-averaged precision, recall, and F1-scores were 78.8%, 79.0%, and 77.5%, respectively. Results show strong discriminative capability but also highlight misclassification challenges among visually similar classes and signs of overfitting. These findings emphasize both the potential and limitations of deep learning when applied to design workflows. The study concludes that CNN-based visual analysis can meaningfully inform design decisions, identify hidden patterns, and support predictive scenario modeling, underscoring the need for interpretability and responsible AI integration in design disciplines.