Table 2 Main results with mean imputationa.

From: Smartphone accelerometer data as a proxy for clinical data in modeling of bipolar disorder symptom trajectory

Features (# of features)

Classifier

Acc

Acc std

AUC

AUC std

Sensitivity

All (50)

Random Forest

0.9450

0.03

0.9912

0.01

0.9391

Gradient Boost

0.9377

0.06

0.9810

0.03

0.9077

Neural Network

0.9275

0.01

0.9751

0.01

0.9844

Typing+Accel (34)

Random Forest

0.9134

0.01

0.9442

0.01

0.9119

Gradient Boost

0.8836

0.02

0.9430

0.02

0.8972

Neural Network

0.8907

0.02

0.9456

0.02

0.9119

Typing (22)

Random Forest

0.8998

0.02

0.9403

0.01

0.9035

Gradient Boost

0.8778

0.02

0.9398

0.01

0.8888

Neural Network

0.8871

0.02

0.9332

0.02

0.8969

Accel (12)

Random Forest

0.8704

0.02

0.9328

0.02

0.9119

Gradient Boost

0.8395

0.03

0.9042

0.02

0.8919

Neural Network

0.8623

0.01

0.9114

0.02

0.8854

  1. aAcc accuracy, AUC area under curve, std standard deviation.