The variety and ambiguity of the hand gestures will have a significant impact on the accuracy and trustworthiness of the recognition task. In the modern age, hand gesture recognition has replaced human interactions. However, in the fields of computer imagination and pattern recognition, detecting the hand component has grown to be a difficult problem. The manual feature extraction approach used in the past is laborious and has a poor recognition rate when it comes to recognizing hand gestures. A new recognition method relying on convolutional neural network (CNN) is suggested in order to increase the recognition rate. Convolutional neural network (CNN) architectures are a highly popular deep learning approach for classification applications, but they have certain drawbacks, including large variation during forecasting, overfitting issues, and prediction mistakes. Nevertheless, proper training a CNN model takes a lot of time. This work presents an ensemble of CNN-based methods to address these issues. A graded ensemble model is provided, which boosts overall network functionality by using the supplementary data supplied by the base model. In this study, deep convolutional feature extraction is carried out using a pre-trained CNN model. Also, the ensembles of machine learning (ML) model are employed as a light weighted modeling over deeper learning (DL) approaches in relationships of lowering the training expense for the image classifications problem employing deep convolutional features. The dataset is used to test the suggested ensemble technique, and the accuracy attained is 99.1%. It has been shown that our suggested ensemble model accomplishes better than other suggested techniques already in use.