Due to the supply problems of fossil-based energy sources, the tendency towards alternative energy sources is relatively high. For this reason, the use of solar energy systems is increasing today. This study combines experimental data and machine learning algorithms to evaluate the energy performance of four different photovoltaic (PV) panel designs (monocrystalline, polycrystalline, flexible, and transparent) under harsh environmental conditions on Horseshoe Island (Antarctica). In this research, the effects of environmental factors, such as solar radiation, temperature, humidity, and wind speed, on the panels were analyzed. Electrical power output of the PV panels are analyzed using six machine learning models. Random forest (RF) and CatBoost (CB) models showed the highest accuracy and reliability among these models. According to the experimental results, Monocrystalline PV provided the highest electrical power (20.5 Watts on average), and Flexible PV provided the highest energy efficiency (19.67%). However, Flexible PV was observed to have higher surface temperatures compared to the other panel types. Furthermore, using Monocrystalline PV resulted in an average reduction of 4.1 tons of CO2 emissions per year, demonstrating the positive environmental impact of renewable energy systems. Thanks to this study, renewable energy research for temporary stations in Antarctica will focus on explainable and interpretable artificial intelligence models that will provide an understanding of the factors affecting the energy performance of PV panels. The research results will be an important guide for optimizing energy consumption, management, and demand forecasting in temporary research stations in Antarctica.