Using sequential patterns as features for classification models to make accurate predictions on ICU events.

Area of Science:

• Clinical informatics
• Data mining
• Machine learning

Background:

• Pattern mining algorithms are used in clinical contexts but generate numerous, often uninterpretable rules.
• Eliciting precise and interpretable patterns is crucial for clinical decision-making, especially in critical care.

Purpose of the Study:

• To present a two-stage sequential contrast patterns based classification framework.
• To enhance the detection of critical patient events, such as hypotension, using interpretable patterns.

Main Methods:

• Utilized a contrast mining algorithm to extract sequential patterns in the first stage.
• Applied post-processing to convert patterns into binary and frequency-based features for classification model development in the second stage.

Main Results:

• The framework demonstrated improved predictive capabilities on eight critical care datasets.
• Sequential patterns, when used as features, enhanced the performance of classification models.

Conclusions:

• The proposed framework offers a more precise and interpretable approach to pattern mining in critical care.
• Sequential contrast patterns are effective features for building advanced learning models to predict critical patient events.