SmartClean-AI: Automated Data and Image Cleaning Libraries for Machine Learning Workflows

Data quality determines the reliability of analysis and machine learning, yet preparing raw data is repetitive and rarely documented. This paper presents SmartClean-AI, a research project that delivers two open-source Python libraries, smartclean-ai for structured data and smartimageclean for image datasets, both published on the Python Package Index under the MIT licence. smartclean-ai accepts a pandas DataFrame, profiles it, handles missing values, removes duplicate rows, and detects and caps outliers with the interquartile-range (IQR) rule, returning the cleaned data, before-and-after profiles and a cleaning report. smartimageclean validates image files, measures blur, brightness and resolution, detects exact and near-duplicate images, computes a dataset quality score, assigns KEEP, REVIEW or REMOVE decisions, standardizes eligible images, re-analyses them and writes CSV and JSON reports. In a recorded run on 45 images, 32 were initially acceptable and 13 were not; after standardization, seven of the 13 satisfied the requirements and six still needed review, and the image library’s 26 automated tests passed. The libraries are rule-based and make every change visible; accuracy against annotated data and comparison with existing tools remain future work.