Skip to main content

Table 3 A summary of the results for testing the effectiveness of the derived diet and physical activity measurements for enhancing obesity estimation using five statistical and machine learning models (i.e., OLS, GWR, RF, DNN, and GRF)

From: Deriving neighborhood-level diet and physical activity measurements from anonymized mobile phone location data for enhancing obesity estimation

  

OLS

GWR

RF

DNN

GRF

City

Fit measures

Base line

Test

Base line

Test

Base line

Test

Base line

Test

Base line

Test

NYC

R2

0.861

0.869

0.975

0.977

0.894

0.898

0.879

0.895

0.934

0.934

RMSE

2.194

2.127

0.926

0.898

1.916

1.881

2.045

1.907

1.508

1.506

adjusted R2

0.860

0.868

0.968

0.969

–

–

–

–

–

–

AIC

8840.7

8723.0

6244.3

6237.8

–

–

–

–

–

–

LA

R2

0.963

0.964

0.972

0.974

0.950

0.950

0.924

0.912

0.951

0.951

RMSE

1.043

1.034

0.903

0.872

1.213

1.210

1.495

1.613

1.204

1.208

adjusted R2

0.962

0.963

0.968

0.970

–

–

–

–

–

–

AIC

2811.5

2800.7

2732.3

2696.6

–

–

–

–

–

–

Buffalo

R2

0.976

0.976

0.982

0.983

0.869

0.873

–

–

0.877

0.875

RMSE

1.088

1.079

0.934

0.914

2.514

2.478

–

–

2.444

2.456

adjusted R2

0.966

0.965

0.969

0.968

–

–

–

–

–

–

AIC

275.4

280.1

272.6

276.8

–

–

–

–

–

–

  1. Adjusted R2 and AIC can only be calculated for the two statistical models; DNN model cannot be trained for Buffalo due to the small number of data records (only 77 data records)
  2. Numbers in bold indicate improvements over the baseline analyses