The widespread use of social media platforms for image sharing has raised concerns about individual privacy. Users often share images without fully understanding the potential privacy risks, which can lead to unintended disclosure of sensitive information. In our paper, we address privacy concerns in social media image sharing by utilizing the PrivacyAlert dataset from Flickr, which is annotated with privacy labels, to differentiate between private and public images. We tackle challenges like dataset bias and limited quantities through data preprocessing, model selection, and fairness assessments. Our study also compares the efficacy of a pre-trained ResNet101 model and a custom CNN model in predicting image privacy. We focus on metrics such as accuracy and fairness, aiming to improve the performance and fairness of image privacy datasets. This research seeks to contribute to safer digital environments by enhancing privacy protection in image sharing.