Demand forecasting is essential for any business to grow and manage its different business activities. With the basic needs of customers, it is hard to predict the future demand using traditional techniques. The popular approach to overcoming the difficulties faced by startup businesses is to use machine learning techniques for demand prediction. Another constraint to train machine learning algorithms for accurate prediction requires a considerable amount of data. For a new startup business, vast data acquisition is a very problematic issue. To overcome the data scarcity problem data enhancement techniques are mainly used for expanding the existing data. Synthetic data generation to balance the existing data which can lead to increased prediction model accuracy. In this study at the beginning, we found that the accuracy of the proposed clustering-based ensemble regression model was bad because of the small size of the data. To overcome this issue, we proposed a modified Conditional Wasserstein Generative Adversarial Network with a Gradient Penalty (CWGAN-GP) for generating synthetic time-series data according to the original data distribution. This generated synthetic data was further added to the original data to train the model and additionally enhanced the demand prediction accuracy for shared electric kickboards. Improved performance was noticed after the model was trained with combined data. Using a range of evaluation measures and graphical representations, we evaluated the performance of our approach against that of other ensemble models. For the production of synthetic data, our GAN model converged more quickly than other GAN models and solved the mode collapse problem. We have contrasted our suggested approach with other cutting-edge models. This study can be helpful for companies to meet the user’s demand for a better quality of service.