ECC-MIMLSVM⁺ is an algorithm of multi-instance and multi-label learning framework. This algorithm proposes a method based on classifier chain, but it does not consider the dependencies between labels. When the number of tags increases, the length of the sub classifier chain also increases, making the error propagation problem prominent. Therefore, this paper presents a kind of improved algorithm, combining ECC-MIMLSVM⁺ algorithm and the label dependencies. ELDCT-MIMLSVM⁺ algorithm is designed, which is based on ensembles of label dependencies classifier chain to avoid the information loss and improve the classification accuracy. The experiment results show that the algorithm has good effect.