In recent years, the accessibility of weather forecasts has reached a point that checking it up on a smart device, like a smartphone, only takes a few seconds. Despite easy accessibility, false predictions of weather forecasts are commonly experienced, which this situation has yet to be put right. In meteorology institutions where this situation arises from, there is a need for many personnel working both on the fields and in the institution to make weather predictions as accurate as possible. As in all human-based systems, human mistakes constantly occur in weather forecasting systems. If human factor was to be minimized in weather forecasting systems, likewise human fallibility would diminish. The most feasible way to deal with this problem is to take advantage of the deep learning techniques, the pinnacle of modern software technologies, which requires almost no human effort on the domain they work, once developed. The deep learning methods have the capability of classifying big image datasets with image processing. With this feature and using the cloud pictures taken from the ground, they can be classified as clear/cloudy and the weather cloudiness can be determined as a numerical ratio. This study aims to reduce human-induced meteorology errors as much as possible with the application of deep learning techniques. The dataset that was preferred contains cloud pictures taken from the ground which are classified as either clear or cloudy. In order to compare different deep learning architectures and their efficiency on this subject, four particular pretrained models were selected. Among the models based on MobileNet V2, VGG-16, ResNet-152 V2, DenseNet-201: VGG-16 came as the best in terms of accuracy with 91.4%. In the future, it can be foreseen that all weather forecasting systems will prefer making their predictions based on modern artificial technologies like deep learning.