AI-Driven Advertisement Optimization Using DenseNet-Swin Transformer Hybrid with Monarch Butterfly Optimization

Good advertisement design is essential in capturing audience attention and increasing user engagement. Yet conventional techniques are dependent on manual adjustments to layout and heuristic-based tactics, which are often inefficient when placing important items like logos, CTAs, and product images. Traditional saliency-based models and CNN-based visual attention prediction techniques are plagued by poor spatial awareness, decreased precision, and inadequate adaptability to varied ad formats, leading to poor brand positioning and lower conversion rates. In this research, a DenseNet-Swin Transformer Hybrid model optimized using Monarch Butterfly Optimization (MBO) is presented to improve visual attention prediction in ads. The innovation of the new method is the combination of DenseNet’s hierarchical feature extraction with Swin Transformer’s global-local attention mechanisms, yielding better spatial feature learning and finer attention mapping. Also, the use of MBO to perform hyperparameter tuning maximizes saliency heatmaps to accurately identify areas of high engagement in advertisements. The resultant hybrid system beats traditional deep learning models by solving some of the key drawbacks of feature extraction, attention drift, and computational cost. The suggested model reaches an AUC of 0.993, outperforming conventional methods, comparative evaluation identifies a click-through rate gain of up to 13.0% and design optimization benefit of 10.5%, evidencing strong performance over conventional methods. The hybrid model is designed to ensure scalability, computational efficacy, and versatility across different ad formats, presenting a robust AI-based solution for digital marketing. This study advances and optimization with AI, opening up the possibility for instant adaptive ad formats, increased audience response, and better marketing effectiveness.