Digital image watermarking is a method for preserving image authenticity and thereby the security of images. Image watermarking is used to embed a watermark image into the host or cover image. Embedding guarantees integrity, so ensuring security for watermarked images. This work presents a hybrid watermarking strategy based on Discrete Wavelet Transform (DWT) followed by Hessenberg decomposition (HD), Singular Value Decomposition (SVD), and Arnold Transform to increase Robustness and Imperceptibility. Using three scaling factors, the approach combines the watermarks of three sizes into the cover images, preserving both the imperceptibility of the cover images and the robustness of the watermarks. The watermark’s resilience against several attacks—including Gaussian noise, JPEG compression, and histogram equalisation is further assessed. The dataset generated from the experimental watermarking results is used for analysis using a machine learning (ML) model. ML analysis revealed the need to identify significant trends and parameter optimisation for enhanced imperceptibility and resilience. PNSR under no attack was shown to be relatively high, with 74.25 dB, according to studies; NC is 1.0 under no attack. This paper presents useful metrics for choosing the optimal value to achieve robust and imperceptible image watermarking, as well as knowledge for the subsequent optimisation. Applying ML integration in the scheme has greatly improved the prediction of which parameter with what value will provide the best balance between robustness and imperceptibility.