|
|
 |
| |
|
|
|
Optimized ML Model with Explainable AI for Threat Detection at Kuwait International Airport |
|
|
|
PP: 1-25 |
|
|
|
doi:10.18576/isl/140101
|
|
|
|
Author(s) |
|
|
|
Abdallah S. Mohamed,
Adel A. Sewisy,
Khaled F. Hussain,
Ahmed I. Taloba,
|
|
|
|
Abstract |
|
|
| Airport security measures have grown in terms of importance, especially in relation to protection of passengers and airport employees, in addition to looking for efficient ways and means of threat identification. Currently, conventional X-ray image processing methods can generate numerous challenges when it comes to detection of concealed threats like weapons and explosives, and other prohibited items likely to compromise security. Threat identification using X-Ray has been a very important factor in the security aspect of airports worldwide including Kuwait International Airport, this study seeks to design a CNN-GRU hybrid model integrated with Firefly Optimization (FO) and Explainable AI (XAI) to enhance threat recognition accuracy in X-ray images among the passengers passing through Kuwait International Airport. Thus, the objective in threat modeling is to improve threat identification while maintaining the system comprehensible to security personnel. The proposed research is innovative because it for the first time presents a framework utilizing CNN to extract spatial features from the images, GRU for modeling temporal dependencies into the images featuring Firefly Optimization for hyperparameter tuning and SHAP for explainability. This strategy proves both high detection rates and, from the operators perspective, observers standpoints, providing the necessary transparency and, therefore, trust to the operating AI models. Moreover, the integration of SHAP-based Explainable AI helps the model to analyse and explain areas of the X-ray images that inform detection of possible threats. Both of these elements, performance and transparency, are important to make the system reliable for its everyday practical use in airports. The experimental results demonstrate that the proposed CNN-GRU-FO-XAI yields accuracy of 99\%, which is higher compared with other models. It also shows the increased precision, recall, and F1-score leaving no doubt that the model has high outcomes and only a few false positives. The SHAP analysis also indicates the most important areas in the X-ray images concerning the threat detection as well as offers more openness and interpretability. This results in almost perfect detection performance and also provides the explanations of the decisions made by the model which is beneficial to increase trust in AI-driven systems and make use of this tool in high-risk environments as airports. |
|
|
|
|
|
|
|
