In last two decades, due to expansion of usage of multimedia especially images and also the trendy image editing tools, a huge amount of altered and forged images has been generated and circulated in social media and world wide web. Forged images are a threat to individuals, organizations, and society in terms of revenue, goodwill, etc. Verifying the authenticity of an image against possible forgery manually is prone to bias and is not feasible. Machine Learning (ML) models have their pros and cons in detecting image forgeries. Deep learning (DL)-based approaches have shown significant potential in the identification of tampered images due to their inherent feature extraction approaches, model configurations, and limitations of a well-distributed, unbiased public dataset. In this work, a fine-tuned pre-trained ResNet50 model has been proposed to detect tampered images. It focuses on detection of copy-move and splicing forgeries. In addition, a single-point crossover and mutation of Genetic Algorithms (GA) are used to address the class imbalance in the CASIA v2 dataset effectively. An extensive evaluation using simulation-based experiments shows that the proposed approach achieves promising and consistent performance with test accuracy and AUC of 0.9202 and 0.9741. A consistent result for both classes along with low computational complexities suggest the effectiveness of the proposed approach over other balancing strategies. The scalable design of the approach improves the reliability of image forgery detection.