Fuzzy Probabilistic Approach to Identify Driving Behavior of Two-Wheeler Drivers Under Mixed Traffic Conditions

The identification and analysis of aggressive driving behaviour are crucial for enhancing road safety and preventing potential hazards on the roads. By closely examining patterns of aggressive behaviour, such as speeding and sudden lane changes, authorities can implement targeted interventions to mitigate risks and implement effective strategies for promoting safer driving habits and reduce the occurrence of aggressive driving incidents. Towards this, the present study comprehensively analyzes aggressive driving behaviour by leveraging the concepts of probabilistic theory and fuzzy set algorithms, which offer a powerful framework for modelling and analyzing uncertain and imprecise data. For the purpose of analysis, the present study collected naturalistic driving data on three fundamental parameters - speed, acceleration, and jerk, from two-wheeler drivers on a 75 km long study stretch in the vicinity of Patna city, India. By establishing relationships between acceleration, jerk, and speed values, the present study developed regression, classification and fuzzy probabilistic approaches for identifying the aggressive driving behaviour. Based on the findings obtained from regression methods, it was noted that both random forest regression and decision tree regression methods demonstrated superior accuracy compared to linear regression. Specifically, linear regression yielded an
$R^{2}$
value of 0.08, indicative of subpar performance. Upon implementation of classification methods, it was observed that both decision tree and random forest classification were performing extremely well on identifying the aggressive behaviour with an accuracy of 99.17%, with an F1-score of 0.88. In order to address the dimensionality challenge, Principal Component Analysis (PCA) is employed to reduce the complexity of the problem, followed by the application of Hidden Markov Models (HMM) on the derived variables. The HMM-PCA approach, although limited by insufficient data, showed the ability to identify non-aggressive patterns accurately and aggressive behaviour with nearly accurate accuracy. From these results, it can be observed that the proposed approach was able to efficiently identify the aggressive driving behaviour. With the help of these insights, specific strategies may be created to lessen the negative consequences of aggressive driving on traffic flow control and road safety, ultimately leading to the construction of a more effective and safe transportation system.