Analysing and detecting extreme-selfie images using ensemble technique

Most individuals, especially young people, are obsessed with taking and sharing selfies online. In the age of TikTok, Facebook, and Instagram, people earn money with exceptional images or videos. However, the competition to attract viewers is not kept safe by the profilers. To some extent, people put themselves in harm’s way to pursue a perfect selfie shot, causing cases of getting hurt or even dying while taking selfies. They take selfies in dangerous locations of mountain peaks, tall buildings, dangerous wild animals, lakes and many other places, which leads to many accidents. Therefore, it is a tentative proposition for the research community to understand the diverse effect of social media. This paper distinguishes between Selfies and Extreme-selfie images to detect risky situations by analyzing the surrounding. We have observed various previous Artificial Intelligence classical techniques in improving automated and accurate solutions for image classification. Additionally, we have used ensemble techniques including VGG16, VGG19, InceptionV3, ResNet50, MobileNetV2, and DenseNet121 models for extreme-selfie identification. It gives predictions based on other algorithms’ results through an average voting classifier method and has shown significant success in classifying extreme-selfie images. Therefore, it outbid all other previous work achieving a validation accuracy of 97.96% and a test accuracy of 98%.