Generalization is a major problem in reinforcement learning (RL), as agents struggle in environments outside their training sets. Often caused by overfitting during the training phase, this issue limits the application of RL in the real world. This paper tries to solve the generalization problem by using a domain randomization technique during the training period. Using two real-world problems; the financial market and agriculture (crop production), this paper trains classical deep reinforcement learning algorithms (DQN and PPO) in different and randomized environments (MDPs). Agents trained in randomized environments generalize better than those trained in single environments (baseline agents). This conclusion is based on the results, in which the agents trained in the randomized environments achieve higher cumulative rewards.