We are excited to announce that our papers “Step-by-Step Data Cleaning Recommendations to Improve ML Prediction Accuracy” and "Icewafl: A Configurable Data Stream Polluter" were accepted at the 28th International Conference on Extending Database Technology (EDBT 2025).
Step-by-Step Data Cleaning Recommendations to Improve ML Prediction Accuracy
Authors
Sedir Mohammed (Hasso Plattner Institute)
Felix Naumann (Hasso Plattner Institute)
Hazar Harmouch (University of Amsterdam)
Abstract
Data quality is crucial in machine learning (ML) applications, as errors in the data can significantly impact the prediction accuracy of the underlying ML model. Therefore, data cleaning is an integral component of any ML pipeline. However, in practical scenarios, data cleaning incurs significant costs, as it often involves domain experts for configuring and executing the cleaning process. Thus, efficient resource allocation during data cleaning can enhance ML prediction accuracy while controlling expenses.
This paper presents COMET, a system designed to optimize data cleaning efforts for ML tasks. COMET gives step-by-step recommendations on which feature to clean next, maximizing the efficiency of data cleaning under resource constraints. We evaluated COMET across various datasets, ML algorithms, and data error types, demonstrating its robustness and adaptability. Our results show that COMET consistently outperforms feature importance-based, random, and another well-known cleaning method, achieving up to 52 and on average 5 percentage points higher ML prediction accuracy than the proposed baselines.
Icewafl: A Configurable Data Stream Polluter
Authors
Christoph Schinninger (opta data Österreich GmbH)
Fabian Panse (University of Augsburg)
Constantin Kühne (Hasso Plattner Institute)
Lisa Ehrlinger (Hasso Plattner Institute)
Abstract
Due to their high velocity, data streams pose additional challenges for analysis and quality assurance compared to static data. To select a suitable computational model to analyze a data stream, or the right data quality tool to clean it, it is important to have benchmark data that reflects the characteristics of the stream such as trends and seasonality, but also potential data errors. Although several data polluters have been developed to inject errors into existing data, none of them supports the creation of temporal data errors as they occur in data streams.
Therefore, we propose Icewafl, a data polluter that allows the injection of temporal errors to create benchmark datasets. Icewafl is built on top of Apache Flink to enable seamless integration with existing data stream pipelines and efficient processing of large-scale data. We show that Icewafl can be used to evaluate (1) the error detection capabilities of data quality tools and (2) the robustness of forecasting methods.
