These results indicate that the proposed framework is capable of reproducing expert-level interpretation of CAS interaction alerts, supporting consistent prioritization aligned with operational safety practices. As shown in Table 1, the obtained sensitivity (recall) was 97.2%, indicating that the system correctly identifies approximately 97 out of every 100 interactions classified as high risk by the CAS specialist. The system also achieved a precision of 90.5%, meaning that most interactions categorized as high risk by the expert system were also interpreted as high-risk situations by the CAS safety specialist. Overall, the framework demonstrates the ability to translate CAS interaction alerts into contextual risk indicators aligned with expert interpretation across a diverse set of operational scenarios. Some prioritized interactions were reconstructed from CAS operational logs using a dedicated visualization tool, as shown in Figure 5, while Table 2 summarizes their operational context and CAS specialist comments. Figure 5 – Schematic reconstruction of representative high-risk interaction scenarios correctly prioritized by the expert system (true positives). Table 2 – Operational context of the scenarios shown in Figure 5. Ite m Zone Aggresso r speed (km/h) Opponen t speed (km/h) Min distanc e (m) System severity (%) CAS specialist comment a) Haul road 25-30 6-9 3-4 ~85 Late LV braking near descending HT. (b) Haul road 18-22 5-10 1-2 ~90 LV remained in HT blind spot. (c) Intersectio n 28-32 5-8 <1.5 ~80 LV entered blind spot with late braking. (d) Loading 24-28 3-5 2-4 ~80 No HT deceleration; LV evasive maneuver. The false negative rate was 2.8%, indicating that a small proportion of interactions interpreted as high risk by the CAS specialist were assigned a lower severity level by the expert system. These cases were primarily associated with zones of low interaction density and operating conditions close to the transition boundaries between severity levels defined by the fuzzy model. Representative examples of these scenarios were reconstructed from CAS operational logs using
RkJQdWJsaXNoZXIy MTM0Mzk2