<p>Optimizing the structure and connections of the SLFN (single hidden layer feedforward neural network) is challenging. ELM (extreme learning machine), known for its non-iterative learning, excels in learning speed and generalization with a fixed network structure. However, it has a drawback of failure to fit into changing network topographies. To address this issue, CGA-ELM (cooperative binary-real genetic algorithm-extreme learning machine) was proposed, using a cooperative genetic algorithm (CGA) to optimize the SLFN structure and parameters simultaneously. Despite its promise, the CGA was implemented without the use of elitism strategy, leading to issues such as premature convergence, reduced population diversity, and high sensitivity to initial population. Moreover, the CGA-ELM has not been statistically evaluated. Therefore, this study introduces an improved CGA with elitism strategy, named ICGA (improved cooperative binary-real genetic algorithm), and integrates it with ELM, forming ICGA-ELM (improved cooperative binary-real genetic algorithm-extreme learning machine). ICGA-ELM aims to achieve a leaner SLFN architecture with enhanced generalization capabilities. A fitness function combining network complexity and training error evaluates the SLFN performance. Binary genetic algorithm optimizes the structural configurations, while a real genetic algorithm and ELM collaboratively fine-tune the network parameters. The classification application’s experimental results confirm the better generalization capabilities of ICGA-ELM than CGA-ELM and ELM. Additionally, the performance of ICGA-ELM was competitive when compared against some state-of-the-art methods, implying that it could be an efficient tool for data classification.</p>