Multiple fiber cores in elastic optical networks have the function of transmitting information, but the distance between the fiber cores is small, which makes it easy to experience crosstalk and increases the error rate of elastic optical networks. To improve the service quality of elastic optical networks and effectively balance the allocation of network resources, a distributed machine learning based elastic optical network resource balance allocation method is proposed. Minimizing the maximum occupied frequency slot number as the optimization objective, establishing a crosstalk optimization model, and introducing the Pisces algorithm in distributed machine learning to solve the model, in order to improve the transmission quality of the fiber core. The resource allocation method of spectrum slicing is used to allocate resources in elastic optical networks, calculate the resource fragmentation rate of links in the network, allocate spectrum time resource windows for services based on the calculation results, and complete resource balancing allocation. The experimental results show that the proposed method has the highest resource utilization rate of 94%, the highest blocking rate of only 2.97%, and the highest energy consumption per unit bit of only 10 nJ/bit, which is practical.