Rapid Urbanization in developing countries is leading to increased congestion and degradation of the environment in metropolitan areas, and Two-wheelers (TW) are contributing substantially to this aspect. TW are becoming popular because they are compact and can weave through traffic easily. On the other hand, TW are more vulnerable because of their direct exposure to the traffic. As per the MoRTH (2022) report, nearly 50% of the accidents on the road are due to TW. In this context, the present study investigates the driving behaviour of TW under Naturalistic Driving conditions. The study aims to understand the impact of road features on TW driving behaviour by selecting 3 different stretches based on their geometric characteristics in Hyderabad City, India. Racelogic Performance Box has been used for the data collection. Various parameters like road geometrics and acceleration and deceleration profiles of the driver have been extracted. Later, the data was analyzed to understand the relationship between velocity and acceleration at various parts of the road sections like curved sections and intersections. In this process, different machine-learning models have been tested. Route 1 with a straight stretch, a linear relationship between velocity and acceleration has been observed with an R2 of 0.8. For Route 2 with curved sections and Route 3 with a combination of curved sections and intersections, the decision tree algorithm was performing well with an R2 of 0.9. As road geometrical elements have a substantial impact on TW driving behaviour and very limited work was reported in the literature the novelty of this study emphasizes on analyzing the impact of these geometrical elements on TW driving behaviour. By exploring the complex relationship between driver behaviour and road environment, the study outcomes can provide insights to policymakers and highway designers for safer design strategies.