LAUSR

Despite all technological advances, runway excursions (REs) remain among the most frequent aviation accidents. Several factors have been shown to contribute to RE incidents. Although different studies have addressed the prediction of REs using quantitative data, the present study contributes to RE research by applying advanced quantitative analysis methods using text mining of voluntarily submitted reports to predict aircraft damage as the outcome of a runway excursion event. The study also intended to unveil relevant predictors from the voluntary safety reports data. The random forest model presented better results than the naïve Bayes and the gradient boosting approaches. The positive predictive value of the random forest method was measured at 74.90%, which is in contrast to 63.58% of the naïve Bayes method and 72.56% of the gradient boosting method. Although several generalizable topics were identified in the reports (such as runway and touchdown zone characteristics, flight control during approach, aircraft malfunction, weather factors, and runway surface and braking action conditions), no defined predictors for runway excursion events resulting in aircraft damage could be established.