Table 3 Performance on the test set with the best set of hyperparameters identified on the validation set. Results of TG are reported as median [25th-75th] percentile. Abbreviations: FP/day, false positives per day; TG, time gain; rAR, recursive Autoregressive model; ARIMA, Autoregressive Integrated Moving Average; NN, Neural Network; LSTM, Long Short-Term Memory Neural Network; CNN-LSTM, Convolutional Long Short-Term Memory Neural Network; RF, Random Forest; LGB, LightGBM
Model | Precision (%) | Recall (%) | F1-Score (%) | FP/day | TG (min) |
|---|---|---|---|---|---|
rAR | 64.38 | 84.43 | 73.06 | 0.17 | 10 [5–15] |
ARIMA | 44.87 | 70.66 | 54.88 | 0.31 | 10 [10–15] |
NN | 64.14 | 55.69 | 59.62 | 0.11 | 10 [10–15] |
LSTM | 68.97 | 59.88 | 64.10 | 0.1 | 10 [10–15] |
CNN-LSTM | 68.70 | 53.89 | 60.40 | 0.1 | 10 [5–10] |
RF | 70.07 | 57.49 | 63.16 | 0.08 | 10 [5–15] |
LGB | 67.52 | 63.47 | 65.43 | 0.1 | 10 [5–10] |