data-driven-svd-regression

Dependencies:

• pandas
• epiweeks

Install with: pip install epiweeks

Columns

Value = totalconfflunewadm This stores -Total Influenza Admissions(Total number of new hospital admissions of patients with confirmed influenza captured during the reporting week (Sunday - Saturday)

Data Cleaning and Missing Values

The influenza hospitalization dataset (analysis_data/target-hospital-admissions.csv) contains a subset of rows with missing values, particularly in the following fields: -location (state identifier) -location_name -value (number of hospital admissions) -weekly_rate These missing values originate from the upstream CDC/Socrata data source and reflect incomplete or non–state-level records rather than data processing errors.

Handling strategy

We do not impute missing values in this dataset. -Rows with missing location or value are excluded prior to population merging and seasonal feature construction, as they cannot be reliably interpreted or linked to state-level population data. -Missing values in weekly_rate are not imputed, since this field is derived and can be recomputed after population data are merged. This approach prioritizes data integrity and aligns with standard epidemiological data-cleaning practices.

Z score normalization

Z-score normalization centers each state's hospitalization time series at zero and scales by its historical standard deviation, enabling comparison of relative flu intensity across states with very different population sizes.

PCA

We constructed a flu matrix with season weeks as rows and state-season pairs as columns, then applied PCA via SVD to both population-normalized and z-score–normalized datasets. We stored the resulting U, Σ, and V matrices in long-form CSVs for full reproducibility.