A scheme based on irregular V-shaped silicon nanoantennas is proposed to optimize transverse unidirectional scattering under plane wave irradiation. Traditional methods of designing regular shapes offer fewer parameters and higher search efficiency. However, due to the limitations of regular shapes, it is challenging to meet high-precision design requirements. Irregular shape design allows for a broader range of adjustments, but the complexity of shape parameters leads to lower search efficiency and a higher likelihood of converging to local optima. This paper proposes an irregular fine-tuning scheme for regular V-shaped silicon nanoscale antennas that combines the advantages of both approaches. Firstly, the regular V-shaped nanoantenna is tuned to generate transverse unidirectional scattering and then transformed into a binary image. Subsequently, the kernel geometry is fixed while the morphology is modified by a surface contour method to form irregular V-shapes. Finally, those irregular V-shapes are input into a pre-trained predictor cascaded by Bayesian optimization (BO) and the nanoantenna's shape is progressively updated by minimizing the mean squared error (MSE) between the target scattering and the predicted scattering of irregular nanostructures. The results demonstrate that the optimized irregular V-shaped nanoantennas exhibit perfect transverse directional scattering and the scattering tail in the opposite direction is greatly shrunk, the MSE of its scattering compared to the ideal unidirectional scattering has decreased by approximately 29% after optimization. Our findings can promote all-dielectric nanoantennas with specific directional scattering in integrated nanophotonics circuits and sensors.