<p>An automated effective and accurate recognition of American sign language (ASL) are the major challenge of moder era. Deep learning-based techniques have made significant progress in ASL recognition, but to achieve enhanced accuracy across different signers and varying environmental conditions is another challenge. To address these issues and to enhance the efficiency of ALS recognition, the proposed method presents an innovative technique which utilizes optimized static ASL recognition with proficient implementation of a weighted average ensemble of three custom-designed convolutional neural network (CNN) models. The proposed method capitalizes on the exclusive strengths of various proposed CNN frameworks, each specifically designed to handle particular nuances of static ASL recognition. The outputs of these custom CNNs are now integrated using bootstrap aggregation (bagging), the resultant ensemble achieved a testing accuracy of 95.53%. Weighted average ensemble technique is used to further refine the proposed architecture which resulted in a notable accuracy improvement to 99.16%. This approach not only represents a substantial improvement in the field of static gesture recognition but also delivers a dependable solution for effective ASL communication. The improved system demonstrates strong potential to enhance accessibility for the hard-of-hearing and deaf community by enabling more precise and efficient interpretation of ASL across a wide range of real-world situations.</p>