The widespread deployment of Internet of Things ( IoT) devices promises substantial benefits, enabling applications ranging from smart homes to connected cities. However, to make IoT truly practical and efficient, several technical challenges must be addressed. These include the limited resources of IoT devices in terms of energy and storage, as well as the heterogeneous communication patterns of IoT applications. These limitations significantly impact data throughput and low-latency requirements imposed by IoT applications such as smart healthcare. Consequently, designing effective congestion control approaches proves to be complex. Our thesis, titled 'Smart Congestion Control Schemes for IoT Networks based on Artificial Intelligence Methods,' proposes leveraging Artificial Intelligence to design innovative congestion control approaches. In our research, we began with a comprehensive review of state-of-the-art congestion control approaches proposed in the literature for IoT networks, comparing them with simple methods based on static congestion control parameters. While straightforward, these approaches can severely impair network efficiency. Our critical analysis of these works highlights their limitations and weaknesses, primarily relying on estimations and network traffic regulations unsuited for IoT applications. In this research, we propose a novel congestion control approach based on Artificial Intelligence (AI). This approach primarily relies on Deep Learning (DL), which infers knowledge from massive and multi-scale data generated by IoT devices to predict network conditions, and Machine Learning (ML), which dynamically adjusts congestion controller parameters. The DL model we propose, based on time series data, predicts congestion based on loss rates. Additionally, we introduce an ML model that dynamically adjusts congestion control parameters based on the loss rates predicted by the DL model. These dynamically predicted congestion control parameters are then assigned to the congestion controller deployed in each IoT device, optimizing throughput, avoiding data loss, and ensuring fairness. In conclusion, our approach, implemented and evaluated using our simulator and the Contiki/Cooja environment, demonstrates that our proposed method outperforms traditional approaches. All quantitative results indicate that our algorithms achieve a better balance between throughput, latency, reliability, overhead, and fairness while meeting the requirements of IoT applications, particularly in patient monitoring scenarios within smart healthcare.