Collective intelligence, defined as the ability of groups to collaborate intellectually and solve complex problems, has become a cornerstone of modern problem-solving frameworks. This paper introduces a novel dynamic aggregation algorithm for collective prediction that balances accuracy, diversity, and computational efficiency. The proposed method dynamically adjusts individual agent weights based on performance and reduces redundancy through diversity-promoting mechanisms. Unlike traditional static or computationally intensive approaches, this algorithm utilizes smoothed prediction errors and correlation penalties to adapt to changing data conditions. Experimental validation on real-world air pollution datasets demonstrates the algorithm’s superior performance compared to conventional methods, particularly in resource-constrained and data-scarce environments. The findings underscore the algorithm’s potential to advance adaptive and scalable solutions in collective intelligence systems.