Since wearing a mask creates respiratory difficulties, different ways of using masks have also emerged. Based on the problem of determining whether the mask is worn correctly, the study aims to determine four different mask-wearing status with machine learning models. A dataset was created, with four-class, that are “masked”, “unmasked”, “masked but under the chin”, and “masked but under the nose”. The dataset includes a total of 2000 images, with 500 images in each class. In order to classify the images, the first step is to extract their features. Within the scope of the study, feature extraction processes were performed through trained convolutional neural network (trained-CNN) models. These models are SqueezeNet, InceptionV3, VGG16 and VGG19. 16 different hybrid models have been proposed in order to perform classification processes with Artificial Neural Network, Logistic Regression, Support Vector Machine models created by using the features obtained from the above-mentioned models and the Stacking Model created by using these models. Out of the hybrid models, the InceptionV3 + ANN model achieved the highest classification accuracy of 91.1%. As a result of the study, a decision support system was proposed to determine the mask-wearing status.