To effectively enhance the classification recognition performance of the extreme learning machine(ELM), an improved artificial ecosystem optimization algorithm(IAEO) with crowding forward-backward and backtracking tactics is proposed, and is employed to optimize the super parameter of the ELM. The crowding forward-backward guidance mechanism is inspired by changes of the number of consumers in an ecosystem, and is constructed by modeling the forward-backward regulating mechanism of the predator-prey relationships between the superior and subordinate organisms. And the local backtracking exploitation strategy inherits the population history optimal information to dynamically re-exploit the local micro neighborhood of decomposers, which contributes to guiding the population evolution and improving the local optimization ability. Numerical results show that two kinds of improved strategies can effectively improve the global exploration and local exploitation performance of the AEO. And the proposed IAEO has higher convergence accuracy, stronger robustness, better high-dimensional optimization applicability. The effectiveness and feasibility of the proposed IAEO algorithm in improving classification performance of the ELM by optimizing its super parameter are showed.