Title: Multidimensional crime prediction technique optimisation combining feature extraction and GAN

Authors: Manna Xie

Addresses: School of Judicial Application, Guangxi Police College, Nanning, 530028, China

Abstract: With rising public safety concerns, effective crime prediction has become critical. Traditional methods struggle with incomplete feature extraction and limited multidimensional data handling. This study proposes a multidimensional crime prediction model integrating feature extraction with generative adversarial networks (GAN). The model achieves deep integration of spatial correlation and temporal dependence through a combined graph convolutional network and transformer architecture. A variational autoencoder optimises the GAN, addressing vanishing gradients and data bias. Experimental results show that after 200 iterations, the model's loss value reaches 0.019, outperforming comparison algorithms (0.036, 0.064, and 0.085). In robbery crime prediction, the model achieves 86.3% accuracy, exceeding the best comparison at 83.2%. These results demonstrate that the proposed model significantly enhances crime prediction performance across multiple crime types, offering an intelligent and efficient forecasting approach.

Keywords: graph convolutional network; GCN; transformer; generative adversarial networks; GAN; VA; multidimensional data; crime prediction.

DOI: 10.1504/IJICA.2026.154193

International Journal of Innovative Computing and Applications, 2026 Vol.15 No.5, pp.289 - 299

Received: 05 Aug 2025
Accepted: 04 Dec 2025
Published online: 16 Jun 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article