Feature tuning—application dependent feature type selection for improved classification accuracy

US11499953B2 · kind B2 · utility

Assignee

• International Business Machines Corporation · US

Inventors

• Aminat Adebiyi · San Jose, US
• Mohammed Abdi · San Jose, US
• Andrea Fasoli · San Jose, US
• Alberto Mannari · San Jose, US
• Luisa D. Bozano · Los Gatos, US

Key dates

• Filing date: Dec 9, 2019
• Grant date: Nov 15, 2022
• Expiry date: May 4, 2041

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N20/00
• WIPO field: Measurement
• WIPO sector: Instruments

Abstract

Provided is a method and system for extracting features from raw data for machine learning processing. Using an array of gas sensors, raw data for at least one compound of interest are extracted based upon the type of output signals. Where the output signals are amplitude-variant, mean features are extracted by chunking the raw data into slices and calculating the mean area under the curve. Where the output signals are amplitude-and-time-variant, mean-plus-slope features are extracted by taking logarithmic values of the raw data and calculating the mean area under the curve.