Energy consumption plays a major role in evaluating the thread machining of cast iron materials. In particular, heat-treated ductile cast iron produced as austempered ductile cast iron (ADI) has mechanical properties such as high strength, ductility and toughness. However, when fitting parts which have thin walls are produced with ADI, the high cooling rates with this method result in loss of strength and fracturing in the walls. This work focuses on a novel study on the effect of response variables in determining energy-power consumption in the thread machining of fitting samples produced with ADI materials under different conditions. The experiments are designed using an L8 orthogonal array. Linear regression is used for developing predictive models and nature-inspired metaheuristic algorithms are employed for optimization. Dragonfly algorithm (DA), a recently developed and popular metaheuristic is considered the primary optimization algorithm. Seven advanced variants of the DA namely, biogeography-based Mexican hat wavelet dragonfly algorithm (BMDA), chaotic dragonfly algorithm (CDA), hybrid memory-based dragonfly algorithm with differential evolution (DADE), hybridization of dragonfly algorithm and artificial bee colony (HDA), hybrid Nelder-Mead algorithm and dragonfly algorithm (INMDA), memory-based hybrid dragonfly algorithm (MHDA) and quantum-behaved and Gaussian mutational dragonfly algorithm (QGDA) are also deployed to investigate the comparative advantage of each in such industrial optimization problem. The results of this study showed that heat treatment and the types of machines are two important response variables in determining energy-power consumption. Among the algorithms, DA outperforms the other algorithms in locating the optima.