01 / Machine learning
Documented workbook outputFilling gaps in survey data.
K-nearest neighbours (KNN), a machine learning method, was used to fill missing closed-ended survey responses. The workbook records each change and checks how well the method reconstructs withheld answers.
Inputs, validation & boundaries
Input: Syndromic Surveillance Data - KNN Imputed.xlsx, including original data, imputed data, audit, and summary sheets.
Method: Distance-weighted KNN with nan-euclidean distance. Ten structural-skip cells were separately rule-coded. Free text and identifiers were preserved.
Validation: The workbook reports 92.9% exact agreement on masked cells for k = 3, compared with 91.3%, 87.4%, and 87.0% for k = 5, 7, and 9. This measures internal reconstruction of missing responses; it is not disease prediction or external clinical validation.
Next step: Reproduce the pipeline in versioned code and evaluate robustness across additional datasets before broader use.