Improvements in the Acquisition and Use of Field Data to Guide Self-Adaptation

[laryferreira]

Problem:

While working with previous field data from COVID-19 Sirio-Libanes ICU Admission, the difficulty in finding good, noise-free and processed data so that we could implement it in the BSN was made clear.
This issue is linked to Gabriel' solution from issue #8

• BSN input data is linked to a very specific form of input, so that real-life cases must be worked through in order to be used, so that perhaps relevant information is discarded for the purposes of the BSN.

General Solution

• Machine Learning for Data Quality: Investigating machine learning techniques to improve the quality of collected data. This can include outlier detection, error correction and automatic sensor calibration.

• Data Filtering: Implement data filters and pre-processing to remove noise and ensure that only relevant data is used in the adaptation.

• Data Standardization: Establish formatting and metadata standards for the field data collected, making it easier to understand and use.

Practical Solution

• Fastest: Rethink data entry for three base cases, as described in Gabriel's issues. (4, 5 weeks implementation)
• Ideal: Implementation of an operant neural network to learn from the various input cases and filter out noise, in order to improve the BSN's performance and remove the mechanical nature of data filtering. (Since I'm no ML expert, I really don't know how long this solution would take, so stick to Gabriel's plan!)