The proper decision-making with contingency predictions of behaviors is an essential ability for maximizing benefits of individuals and societies. However, the neural basis of such proper decision-making underlying behavioral selections under conflicting predictions of reward and punishment remains unknown. To investigate the neuronal underpinnings of the adaptive behavioral selections, we trained male Long-Evans rats with an operant discrimination task that includes decision-making for reward only or reward followed by footshock punishment while local field potentials in multiple brain regions were recorded. Rats were subjected to discriminate 9 kHz and 4 kHz pure tones in Go or NoGo trials for reward only or reward followed by footshock (0.2 - 0.5 mA for 1 s), respectively. A semi-supervised machine-learning technology called discriminative cross-spectral factor analysis successfully extracted an oscillatory brain activity pattern that decreases several seconds before the reward-taking behaviors in the NoGo trials happened; we named the oscillatory brain activity pattern as the careful decision-making pattern. The careful decision-making pattern included beta and gamma oscillations in the prefrontal cortices and beta coherence between the medial prefrontal cortex and the amygdala. Chemomgenetic inhibition of the beta and gamma oscillations in the prefrontal cortices and the beta coherence between the medial prefrontal cortex and the amygdala reversibly impaired the proper behavioral selections between Go and NoGo trials. These results suggest a top-down control of impulsive decision-making by the prefrontal cortices on the subcortical limbic structures including the amygdala.