Götte, Gesa MarieGesa MarieGötteBaier, SimonSimonBaierEhrhardt, InaInaEhrhardtHerzog, AndreasAndreasHerzogZiesak, MartinMartinZiesak2026-03-142026-03-1420249798350355444https://publica.fraunhofer.de/handle/publica/51075510.1109/MetroAgriFor63043.2024.109488102-s2.0-105003533970Forest road networks are vital for forest management and timber transport, yet their maintenance is costly and traditionally reliant on subjective visual inspections. This study introduces a machine learning (ML) based approach combined with a multi-sensor system for automated forest road condition classification. The system, developed within the Intelliway project, utilizes ultrasonic sensors, gyroscopes, accelerometers, and GNSS units mounted on a vehicle to gather comprehensive road data. Data from these sensors are processed using ML techniques to classify road sections into four condition categories. Our methodology includes rigorous data preprocessing, feature extraction, and the application of a decision tree ensemble model optimized for imbalanced multi-class training. The model achieved an overall accuracy of 94.3%, with specific accuracies of 98% for RC1, 90% for RC2, 99% for RC3, and 94% for RC4. Feature importance analysis highlighted that the most crucial inputs were from the acceleration sensors, indicating road roughness as a key condition indicator. The practical implications of this work are significant for forest road managers, offering a cost-effective, efficient, and accurate method for road condition assessment. By automating the monitoring process, the system reduces labor and enhances the consistency of assessments, thus optimizing maintenance planning and lowering long-term repair costs. This technology promises to support more sustainable forest road management practices by ensuring better resource allocation and timely maintenance interventions.enfalseForestroad quality classificationmachine learning and AIroad managementsensorsconference paper