Feature tuning—application dependent feature type selection for improved classification accuracy
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.
1. A method comprising the steps of:
(a) obtaining at least one compound of interest for testing on at least one gas sensor;
(b) obtaining multiple output signals from the at least one gas sensor for the at least one compound of interest;
(c) determining whether each of the multiple output signals is an amplitude-variant signal or an amplitude-and-time-variant signal; and
(d) for any amplitude-variant output signal, extracting its mean features, and for any amplitude-and-time-variant output signal, extracting its mean-plus-slope features,
wherein,
mean feature extraction is performed on any amplitude-variant output signals by chunking each of the amplitude-variant output signals into slices and calculating mean area under the curve, wherein values for all of the slices represent the curve for the amplitude-variant signals, and
mean-plus-slope feature extraction is performed on any amplitude-and-time-variant output signals by taking logarithmic values for each of the amplitude-and-time-variant output signals and calculating mean area under the curve, wherein all of the logarithmic values represent the curve for the amplitude-and-time-variant output signals.
2. The method of claim 1 , comprising:
(e) combining the extracted mean features and/or the extracted mean-plus-slope features to classify the at least one compound of interest.
3. The method of claim 1 , wherein the at least one compound of interest is a volatile organic compound (VOC).
4. The method of claim 1 , wherein the at least one compound of interest comprises at least two compounds of interest and the amplitude-variant output signals indicate that the at least two compounds of interest share similar functional groups and emit volatiles with similar chemical groups.
5. The method of claim 1 , wherein at least one compound of interest is at least two compounds of interest and the amplitude-and-time-variant output signals indicate that the at least two compounds of interest have different functional groups and emit volatiles with different chemical groups.
6. The method of claim 1 , wherein the at least one gas sensor is part of a gas sensor array.
7. The method of claim 1 , wherein the extracted features are formatted for machine learning processing.
8. A method for fine-tuning features from raw data comprising:
(a) obtaining multiple output signals for at least one compound of interest;
(b) determining if the multiple output signals are amplitude-variant or amplitude and time-variant through visualization of the multiple output signals; and
(c) extracting features from the raw data for the multiple output signals,
wherein,
mean features are extracted on the raw data from the multiple output signals that are amplitude-variant by chunking the amplitude-variant raw data into slices and calculating mean area under the curve, wherein values for all of the slices represent the curve v, and
mean-plus-slope features are extracted on the raw data from the multiple output signals that are amplitude-and-time-variant by taking logarithmic values of the amplitude-and-time-variant raw data and calculating mean area under the curve, wherein all of the logarithmic values represent the curve for the amplitude-and-time-variant output signals.
9. The method of claim 8 , wherein the multiple output signals are obtained from a gas sensor array.
10. The method of claim 8 , wherein the at least one compound of interest is a volatile organic compound (VOC).
11. The method of claim 8 , wherein the at least one compound of interest is at least two compounds of interest and the amplitude-variant output signals indicate that the at least two compounds of interest share similar functional groups and emit volatiles with similar chemical groups.
12. The method of claim 8 , wherein the at least one compounds of interest is at least two compounds of interest and the amplitude-and-time-variant output signals indicate that the at least two compounds of interest have different functional groups and emit volatiles with different chemical groups.
13. The system of claim 12 , wherein the at least one algorithm formats the extracted features for machine learning processing.
14. The system of claim 12 , wherein the at least one compound of interest is a volatile organic compound (VOC).
15. The system of claim 12 , wherein the at least one compound of interest is at least two compounds of interest and the amplitude-variant output signals indicate that the at least two compounds of interest share similar functional groups and emit volatiles with similar chemical groups.
16. The system of claim 12 , wherein the at least one compound of interest is at least two compounds of interest and the amplitude-and-time-variant output signals indicate that the at least two compounds of interest have different functional groups and emit volatiles with different chemical groups.
17. The method of claim 8 , wherein the extracted features are formatted for machine learning processing.
18. A system comprising:
a sensor array comprising a plurality of gas sensors for testing at least one compound of interest, wherein output signals from each sensor of the sensor array are amplitude-variant output signals and/or amplitude-and-time-variant output signals; and
a microprocessor comprising at least one algorithm in communication with the sensor array, wherein the microprocessor applies mean feature extraction on any amplitude-variant output signals and mean-plus-slope feature extraction on any amplitude-and-time-variant output signals.
19. The system of claim 18 , wherein the microprocessor extracts mean features from the output signals by chunking values of the amplitude-variant output signals into slices and calculating mean area under the curve, wherein values for all of the slices represent the curve for the amplitude-variant output signals.
20. The system of claim 18 , wherein the microprocessor extracts mean-plus-slope features from the output signals by taking logarithmic values of the output signal values and calculating mean area under the curve, wherein all of the logarithmic values represent the curve for the amplitude-and-time-variant output signals.
21. A system comprising:
a sensor array comprising a plurality of gas sensors for testing at least one compound of interest, wherein output signals from each sensor of the sensor array are amplitude-variant output signals; and
a microprocessor comprising at least one algorithm in communication with the sensor array, wherein the microprocessor applies mean feature extraction on the output signals by chunkin values of the amplitude-variant output signals into slices and calculating mean area under the curve, wherein values for all of the slices represent the curve.
22. The system of claim 21 , wherein the at least one algorithm formats the extracted features for machine learning processing.
23. The system of claim 21 , wherein the at least one compound of interest is a volatile organic compound (VOC).
24. The system of claim 21 , wherein the at least one compounds of interest is at least two compounds of interest and the amplitude-variant output signals indicate that the at least two compounds of interest share similar functional groups and emit volatiles with similar chemical groups.
25. A system comprising:
a sensor array comprising a plurality of gas sensors for testing at least one compound of interest, wherein output signals from each sensor of the sensor array are amplitude-and-time-variant output signals; and
a microprocessor comprising at least one algorithm in communication with the sensor array, wherein the microprocessor applies mean-plus-slope feature extraction on the output signals.
26. The system of claim 25 , wherein the microprocessor extracts mean-plus-slope features from the output signals by taking logarithmic values of the output signal values and calculating mean area under the curve, wherein all of the logarithmic values represent the curve.
27. The system of claim 25 , wherein the at least one algorithm formats the extracted features for machine learning processing.
28. The system of claim 25 , wherein the at least one compound of interest is a volatile organic compound (VOC).
29. The system of claim 25 , wherein the at least one compounds of interest is at least two compounds of interest and the amplitude-and-time-variant output signals indicate that the at least two compounds of interest have different functional groups and emit volatiles with different chemical groups.