M.S. Applied Data Science - Capstone Chronicles 2025

17

Figure 15 Fluoride Forecast in the next 10 years

The Fluoride forecast exhibits a gradual upward trend in median concentrations from approximately 0.8 µg/L in 2025 to about 2.0 µg/L by 2035 accompanied by modest seasonal variability (slightly higher in late winter/early spring). Historical scatter shows many spikes up to ~25 µg/L, but the SARIMAX model’s predictions remain firmly within the 95 % confidence band of 0–4 µg/L. This suggests that while fluoridation dosing practices in many California systems are relatively stable, slight increases may occur over time potentially reflecting gradual changes in source ‐ water blending or adjustments to maintain optimal dental health targets (typically 0.7 mg/L or 700 µg/L). Because the U.S. Public Health Service recommends fluoride levels of 0.7–1.2 mg/L (700–1,200 µg/L) for dental protection, our projected values remain well below those targets, indicating that most systems may be under ‐ dosing. The narrow confidence intervals likewise imply consistent operational control. Water agencies should consider these forecasts when setting fluoridation parameters and planning public health outreach to ensure compliance with optimal fluoride levels and to

guard against both under- and over-fluoridation risks.

= ∑ ​∑ ​ ​ ​− ˉ ( )( ​− ˉ) ∑ ​( ​− ˉ)2 ​ Yt=Tt+St+Rt 1− β0 + =1 ∑ β =

(1)

(2)

(3)

0 1 ∫ ( −1 ( )

ROC-AUC:

(4)

Before modeling, we compute Moran’s I on county ‐ level annual means to test whether contaminants like arsenic or manganese are randomly distributed across California or exhibit regional “hot spots.” High Moran’s I suggest we should incorporate spatial lags or geographically weighted regressions in predictive models. (1) We apply this decomposition to each county–analyte series to visualize whether there is a rising baseline (perhaps due to aging infrastructure) or a consistent seasonal pattern (e.g., dilution in the rainy season). (2) We use logistic regression as a baseline “two-stage” approach: first classify whether a

89