Neural Network-Based Train Identification in Railway Switches and Crossings Using Accelerometer Data

Table 6

Mean accuracy (with standard deviation if applicable) for different neural network models and different scenarios. Baseline (random) accuracy is denoted as “Base.” Visualization of this table is shown in Figure 5.

Model/Scenario	Base (%)	Mean (%)	MLP1 (%)	MLP2 (%)	MLP3a (%)	MLP3b (%)	CNN (%)	LSTM (%)
A	14.3	52.3 ± 7.9	50.9 ± 4.4	52.3 ± 8.2	51.4 ± 7.0	49.5 ± 8.2	60.0 ± 6.3	49.7 ± 7.2
B	20.0	80.6 ± 12.0	82.9 ± 5.7	87.3 ± 7.0	83.3 ± 6.0	81.2 ± 6.0	89.2 ± 6.9	59.8 ± 9.7
C	20.0	72.9 ± 9.9	76.2 ± 7.1	74.1 ± 5.2	73.7 ± 9.2	73.5 ± 4.9	80.6 ± 6.9	59.3 ± 9.8
D	20.0	55.0	57.5	58.8	53.7	53.1	72.5	34.4