Background: The deep learning algorithm is the machine learning technique where the computers can learnfrom the dataset. This system builds a patterns and models from a given dataset. Expert system machinelearning system and deep neural networks are considerably more efficient in detecting a cancer cell and alsothe nature of the cancer. Cancer prediction and identification are the challenge of our time. Cancer is theleading cause of human death. It is a problem worth solving using available technology. Methods: To train thedatabase to classify and predict acute lymphoma leukemia with minor errors and low computational time weused round robin classification model takes a dataset with three subfolders which are Testing data (Containsboth healthy and leukemia cells), training data (Contain in separate subfolders of healthy and leukemia cells),validation data (folder where the results will be saved). All the samples are in image format. All the image arein .bmp format and contain a resolution of 450x450 pixels and a 24 bit color display system. All the image dataare converted into NumPy array. We will split all this data into train, test and validation purpose. After that, wetrain our proposed sequential CNN Model with our dataset. In our sequential model, we used tensorflow, keras,Dense, Activation, Dropout, Conv2D, numpy, pandas, shutil, time, Cv2, tpdm, sklearn, matpoltlid, seaborn,confusion matrix. Results: We made a custom dataset which we need then to examine confusion metrics, totaltest accuracy, precision rate, recall rate, computational time, F1-score, and MSE (Mean Squared Error) value,and error rate. We also employed model performance with train, validation accuracy, and loss for graphicaldepiction. We successfully achieved significant improvements in our model accuracy. Where other researchersproposed CNN models accuracy given in the accuracy table, the CNN Model accuracy was 93% and CNNModel was 92.97% accuracy on the other hand our proposed model got 78.46% accuracy and when we tried toimprove our model then we got 97.19% accuracy. Conclusion: In every experiment, we used a different kind ofstep to increase the model performance for our dataset. That gave us a high accuracy rate, and the validationloss was low, and testing accuracy was also increased. The medical experts can benefit from the improvedapplication of deep learning, which ensures less time and enhance reliable healthcare for patients. Machinelearning will bless for the medical sector in the near future. Collecting relevant medical data for training amachine learning model is insufficient to do such research.