Classical handwriting recognition algorithms relied heavily on historical data and individual traits. Training an OCR system is challenging when these factors are considered. Deep learning-based handwriting recognition research has advanced significantly in recent years. More research and identification accuracy are required as the volume of handwritten data increases and computer power becomes more readily available. Convolutional neural networks (CNNs) are the most effective technology for solving handwriting recognition difficulties because they automatically extract information from handwritten letters and words. The proposed research will concentrate on convolutional neural network-based handwritten digit detection. We will look at stride size, receptive field, kernel size, padding, and dilution. Our second aim is to determine how SGD optimisation strategies improve handwritten digit recognition accuracy. Ensemble design may improve network recognition accuracy. Ensemble topologies make testing more difficult and raise computational costs, thus we employ a pure CNN architecture to get the same accuracy. The goal of this research is to develop a convolutional neural network (CNN) architecture that can reduce operating procedure complexity and cost while enhancing accuracy over ensemble techniques.