Pipeline-Driven Handwritten Digit Recognition: A ZenML Framework Approach for Efficient Machine Learning Workflows

The integration of machine learning processes has been crucial for creating a scalable and replicable process in many areas. In this work, we present a pipeline-based architecture for solving handwritten digit recognition, implemented with ZenML. In the proposed system, for effective digit recognition of the MNIST dataset, a Support Vector Classifier (SVC) is used as the classifier algorithm to get high accurate result. Using ZenML the identified main steps of data preprocessing, model training, model evaluation and visualization are encapsulated as traceable steps in a workflow. Major input includes having a substantially sound visualization step to display sample prediction and a confusion matrix to allow interpretability and correctness analysis. This framework has a dashboard that helps to keep tabs on the pipeline and enables easy amplification of all the experiments and added capability. This work shows how ZenML manages to provide a capable and flexible solution to deal with the difficulties of ML pipelines trying to be both easy to use and reproducible while also being a substantial tool to support data-scientists and researchers for data-oriented tasks achieving great results.