In Wireless Sensor Network, there are many tiny sensors that plays vital role in various technologies and cluster head selection is an important issues for data transmission. As cluster heads receives data from sensor nodes and sends data after aggregation to the base station. Appropriate cluster heads selection leads to efficient energy consumption and enhances network lifetime. For solving this issue many solutions has been given considering some attributes but with time meta-heuristics algorithms were now widely used for real world applications. Nature inspired algorithm has seeking wide attention of researchers as it gives the capability to self-learn and has better performance. In this paper we have proposed Dragon Fly algorithm which is inspired by dynamic and static behavior of dragon fly. We have first use LEACH for energy model and also compute multi-attributes of sensor nodes using MADM (multi-attributes decision making) accordingly. After this, then the proposed algorithm NDFMA is applied where Dragon fly algorithm and MADM used for optimized selection of cluster heads. The proposed algorithm has been evaluated in terms of numbers of alive nodes, dead nodes, energy consumption, throughput and number of packets sends. NDFMA method suggested that it performs better than other state-of-the art algorithms. The proposed method is way to compute multi-attributes of sensor nodes for ranking them and selecting optimized cluster heads. NDFMA algorithm has been compared with NBA, LEACH and ESO-LEACH which validates that the new proposed algorithm is performing better than the classical and compared algorithm.