Saleem, Mohammed Talal (2023) Driver behaviour classification framework in the context of car following for drivers in Malaysia based on random forest algorithm. Doctoral thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p MOHAMMED TALAL SALEEM.pdf Download (8MB) | Preview
	Text (Copyright Declaration) MOHAMMED TALAL SALEEM COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (364kB) | Request a copy
	Text (Full Text) MOHAMMED TALAL SALEEM WATERMARK.pdf Restricted to Registered users only Download (6MB) | Request a copy

Abstract

Increasing road fatalities year after year is a significant challenge in many countries. However, real-time driver behaviour modelling in the car-following context for drivers in Malaysia has not been adequately studied due to issues with dataset, including availability, accuracy, completeness, diversity, data acquisition system availability, and techniques suitability. Hence, a five-phase methodology was designed to model driver behaviour in the car-following context for drivers in Malaysia. Review of previous work from literature, designing of a data collection system (DAS), profiling and analysing driver behaviour in the car-following context, classifying driver behaviour in the car-following context and solving issues related to missing LiDAR data represents the five phases of methodology, respectively. The real-time experiment on a federal highway road in Malaysia involved 30 participants. The results indicated that the proposed DAS design's efficiency, power, and capacity enabled it to collect driver behaviour data, pre-process it (on-board), and gather missing information due to road curvatures. Also, gender and age have statistically significant effects on drivers’ behaviour during the car-following context for drivers in Malaysia. Female drivers are speedier (risky) and more cautious (keeping a long distance to the front car) than their male counterparts. Female drivers produced the optimum style and good stability behaviour in traffic flow and reported efficient driving style, while traffic status evaluation is difficult to be predicted when having male drivers on road. Senior drivers are regarded as safe drivers (less speedy) and are more prone to causing traffic disturbances than other groups of drivers. Middle-aged drivers reported efficient traffic flow characteristics. The classification framework showed that the Random Forest algorithm is the best technique, achieving 100% accuracy, 100% precision, and 100% recall, with speed and distance as the most effective classification features. The Random Forest algorithm can also impute the missing values of the LiDAR sensor. The result confirmed that the classification accuracy did not vary when using an imputed dataset (validation for the imputation process)

Actions (login required)

View Item