The Internet of Things (IoT) has become a paradigm shifter, enabling digital systems and a vast number of physical devices to be seamlessly interconnected for better automation and real-time decision-making. Antenna is the crucial part of the IoT system, which facilitates short-distance and long-distance wireless communication between devices and networks. Low-power wide area network (LPWAN) technology has special attention as it provides long-distance communication up to 10 km with low power, low bit rate, and low cost. Among LPWAN protocols, long-range wide area network (LoRaWAN) protocols are popular, which operate in unlicensed frequency bands, i.e., 433, 868, and 915 MHz. This paper reviews both single-band and multi-band traditional Long Range (LoRa) Microstrip Patch Antennas (MPAs) and explores recent advancements made with Machine learning (ML) techniques. Traditional antenna design often faces challenges in balancing size, bandwidth, gain, and efficiency, while the fine-tuning process is typically time-intensive. Machine learning presents a promising alternative, enabling data-driven design methods that optimize parameters, make adjustments, and predict performance more efficiently. Machine Learning strength lies in its ability to manage complex design parameters, address nonlinearities, and tackle multi-objective optimization tasks, leading to faster design times and greater accuracy. This paper concludes that ML has significant potential to advance LoRa MPA design by improving efficiency, speeding up the design process, and enhancing overall performance.