Explainable analytics: understanding causes, correcting errors, and achieving increasingly perfect accuracy from the...

Area of Science:

• Machine Learning
• Data Science
• Predictive Analytics

Background:

• Accurate data analytics are crucial, but understanding the sources of inaccuracy is equally important.
• Existing methods often lack transparency, making it difficult to trace prediction errors.

Purpose of the Study:

• To introduce a Transparent Classification (TC) method for enhanced data analysis.
• To provide a traceable and correctable approach to prediction errors.
• To identify potential inaccuracies in ground truth data.

Main Methods:

• TC utilizes positive and negative patterns derived from training data intersections.
• Pure positive and negative patterns are identified for classification.
• Observations are scored and classified based on these patterns.

Main Results:

• TC successfully identifies all positive observations with minimal training data (e.g., 1:9 ratio for Breast Cancer Wisconsin).
• The method eliminates uncertainty by avoiding parameter tuning and random selection.
• TC visualizes error causes, enabling traceability and correction.

Conclusions:

• TC offers a transparent and effective method for classification tasks.
• The approach enhances prediction accuracy and aids in identifying diagnostic errors.
• TC's ability to trace and correct errors makes it a valuable tool in data science.