UCI-SECOM: Semiconductor Manufacturing Process Data with 591 Features

Data from a modern semiconductor manufacturing process, collected via sensors and process measurement points for monitoring. The dataset contains 1,567 production examples, each with 591 measured features and a pass/fail yield label, with 104 recorded failures. It was donated to the UCI repository in 2008 for tasks like classification and causal discovery.

Use Cases

• Ranking manufacturing process features for predictability based on their impact on product yield.
• Building classification models to predict pass/fail outcomes based on sensor and measurement data.
• Investigating causal relationships between process signals and downstream yield excursions.
• Applying feature selection techniques to identify the most relevant signals from a large set of 591 variables.

Strengths

• Contains 1,567 real-world production examples, providing a substantive sample for analysis.
• Includes 591 attributes, offering a high-dimensional feature space for feature selection research.
• Explicitly documents 104 fail instances, enabling work on imbalanced classification and anomaly detection.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Last update date is unknown; freshness unverified as the data is from 2008.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

UCI Machine Learning Repository via OpenML.

Collection Method

Likely collected from sensors and measurement points in a semiconductor fabrication facility.