PURPOSES : To prevent an increasing number of drowsiness-related accidents, considering driver fatigue is necessary, which is the main cause of drowsiness accidents. The purpose of this study is to propose a methodology for selecting drowsiness hotspots using continuous driving time, a variable that quantifies driver fatigue. METHODS : An analysis was conducted by dividing driver fatigue, which changes according to time and space, into temporal and spatiotemporal scenarios. The analysis technique derived four evaluation indicators (precision, recall, accuracy, and F1 score) using a random forest classification model that is effective for processing large amounts of data. RESULTS : Both the temporal and spatiotemporal scenarios performed better in models that reflected the characteristics of road sections with changes in time and space. Comparing the two scenarios, it was found that the spatiotemporal scenario showed a difference in precision of approximately 10% compared with the temporal scenarios. In addition, [Model 2-2] of the spatiotemporal scenario showed the best predictive power by assessing the model’s accuracy via a comparison of (1-recall) and precision. This shows better performance in predicting drowsy accidents by considering changes in time and space together rather than constructing only temporal changes. CONCLUSIONS : To classify hotspots of drowsiness, spatiotemporal factors must be considered. However, it is possible to develop a methodology with better performance if data on individuals driving vehicles can be collected.

Headwater streams provide various microhabitats, resulting in high diversity of macroinvertebrate community. In this study, we compared the differences of communities between two adjacent headwater streams (Jangjeon stream (GRJ; GRJ1-GRJ5) and Haanmi stream (GRH; GRH1-GRH3)) in Jungwang and Gariwang mountains, Gangwon-do and evaluated the effects of habitat condition to the macroinvertebrates community composition. In order to characterize the macroinvertebrate communities and extract influential environmental factors, we applied to Cluster analysis (CA), Indicator species analysis and Non-metric multidimensional scaling (NMDS). Total 33,613 individuals in 3 phyla, 5 classes, 13 orders, 51 families, and 114 taxa (genera or species) were collected. Gammarus sp. was dominant at the upper stream of GRJ, whereas Chironomidae spp. was abundant at GRH and the downstream of GRJ. The CA classified samples into six clusters (1-6) reflecting spatial and temporal variation of benthic macroinvertebrate communities. Benthic macroinvertebrate community composition was significantly different between two adjacent streams. Sweltsa sp. 1, Psilotreta kisoensis, Rhyacophila shikotsuensis and Serratella setigera were identified as representative indicator species for clusters 1, 2, 3 and 5, respectively. Similar to CA results, NMDS revealed the spatial and temporal differences of benthic macroinvertebrate communities, indicating the difference of community composition as well as microhabitat condition. Forest composition, proportion of boulders (>256 mm), and water velocity were main factors affecting the macroinvertebrate community composition.