Acute lymphoblastic leukaemia (ALL) is a blood malignancy that mainly affects adults and children. This study looks into the use of deep learning, specifically Convolutional Neural Networks (CNNs), for the detection and classification of ALL. Conventional techniques for ALL diagnosis, such bone marrow biopsy, are costly and prone to mistakes made by hand. By utilising automated technologies, the research seeks to improve diagnostic accuracy. The research uses a variety of pre-trained CNN models, such as InceptionV3, ResNet101, VGG19, DenseNet121, MobileNetV2, and DenseNet121, to extract characteristics from pictures of blood smears. ANOVA, Recursive Feature Elimination (RFE), Random Forest, Lasso, and Principal Component Analysis (PCA) are a few of the selection approaches used to find the most relevant features after feature extraction. Following that, machine learning methods like Naïve Bayes, Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbours (KNN) are used to classify these features. With an 87% accuracy rate, the ResNet101 model produced the best results, closely followed by DenseNet121 and VGG19. According to the study, CNN-based models have the potential to decrease the need for medical specialists by increasing the speed and accuracy of ALL diagnosis. To improve model performance, the study also recommends expanding and diversifying datasets and investigating more sophisticated designs such as transformers. This study highlights how well automated deep learning systems do medical diagnosis.