Traffic accident classification using IndoBERT

Abstract

Traffic  accidents  are  a  widespread  concern  globally,  causing  loss  of  life, injuries,  and  economic  burdens.  Efficiently  classifying  accident  types  is crucial   for   effective   accident   management   and   prevention.   This   study proposes   a   practical   approach   for   traffic   accident   classification   using IndoBERT,   a   language   model   specifically   trained   for   Indonesian.   The classification task involves sorting accidents into four classes: car accidents, motorcycle   accidents,   bus   accidents,   and   others.   The   proposed   model achieves  a  94%  accuracy  in  categorizing  these  accidents.  To  assess  its performance,  we  compared  IndoBERT  with  traditional  methods, random forest (RF) and support  vector  machine (SVM),  which  achieved  accuracy scores   of   85%   and   87%,   respectively.   The   IndoBERT-based   model demonstrates its effectiveness in handling the complexities of the Indonesian language,  providing  a  useful  tool  for  traffic  accident  classification  and contributing to improved accident management and prevention strategies.