ADVANCEMENTS IN AIRLINE SECURITY: EVALUATING MACHINE LEARNING MODELS FOR THREAT DETECTION

Аннотация

INTRODUCTION Vector Machine (SVM), and Neural Network The rapid growth of the airline industry has made (NN)—to determine their effectiveness in security a critical concern, particularly with the predicting potential airline security threats. By increasing number of passengers, flight routes, and analyzing these models' performance based on key baggage passing through airports daily. Ensuring metrics like accuracy, precision, recall, and F1- the safety of passengers, staff, and facilities has score, the research identifies the most suitable ML necessitated the adoption of more sophisticated model for real-world deployment in airline and automated systems to assess potential threats. security systems. Traditional security measures, such as manual screenings and checklists, have proven insufficient The application of machine learning in airline in addressing the complexities and scale of modern security has garnered increasing attention as the airline security. As a result, machine learning (ML) industry seeks to improve risk management and models have become vital tools in improving the threat detection. Traditionally, airline security accuracy and efficiency of security screening systems have relied on rule-based frameworks processes. that follow predefined protocols for screening passengers and baggage. However, such systems Machine learning offers powerful solutions for often suffer from inefficiencies due to their identifying potential security risks by analyzing inability to adapt to new types of security risks, large datasets, including passenger demographics, especially with evolving technologies and tactics travel histories, baggage details, and behavioral used by malicious actors (Kumar & Shankar, patterns. The ability to automate and enhance 2017). decision-making through predictive algorithms can significantly reduce human error and LITERATURE REVIEW streamline security procedures, ensuring a safer travel experience for all. This study aims to Machine learning, particularly supervised learning evaluate four popular machine learning models— algorithms, offers a promising alternative by Decision Tree (DT), Random Forest (RF), Support automatically learning from historical data and https://www.theamericanjournals.com/index.php/tajet 87 THE USA JOURNALS THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN – 2689-0984) VOLUME 06 ISSUE10 detecting patterns that are difficult for traditional their ability to model complex, non-linear systems to capture. Decision Trees (DT), for relationships in large datasets. Unlike traditional example, have been used for their simplicity and machine learning models, Neural Networks can ease of interpretation, making them a preferred automatically learn intricate patterns from data choice in domains where explainability is critical without relying on manually designed features, (Loh, 2011). Despite their interpretability, which makes them highly adaptable for large-scale however, DT models often struggle with overfitting systems like airline security. However, they are and fail to generalize well when dealing with often criticized for their “black-box” nature and complex, non-linear relationships in data, as computational cost (LeCun et al., 2015). Despite shown in recent security research (Breiman, these challenges, recent research indicates that 2017). Neural Networks outperform traditional algorithms in high-dimensional data analysis, Random Forest (RF), an ensemble learning making them an attractive option for detecting technique that builds multiple decision trees, has potential security threats (Goodfellow et al., 2016). been proposed as a more robust solution for airline security challenges. By averaging predictions In conclusion, the literature highlights the across several decision trees, RF reduces the risk strengths and weaknesses of various machine of overfitting and improves overall model learning models in airline security applications. performance. Breiman (2001) demonstrated the While Decision Trees offer simplicity and effectiveness of Random Forest in handling large interpretability, they tend to overfit complex data. datasets and noisy data, making it particularly Random Forest improves generalization through suitable for high-stakes applications such as airline an ensemble approach but requires more security. Recent studies have further validated computational resources. Support Vector RF’s capability to generalize better than individual Machines are effective for clear class separations decision trees while offering high accuracy in but are computationally intensive. Neural classification tasks (Zhang et al., 2019). Networks show exceptional performance in handling non-linear patterns, but their complexity Support Vector Machines (SVM) have been widely and high computational demands pose challenges used in binary classification problems, including for real-time implementation. This study builds on threat detection, due to their ability to find t