Table 9 presents the R2 and mean absolute percentage error (MAPE) values for the training and test datasets. R2 provides a useful and intuitive representation of the model strength, while MAPE indicates the average size of the absolute errors as percentages of the observed 
-values. The test dataset values represent the model performance on data which were not used to train the model in its development. The R2 values were considered to be good, exceeding 0.7 for both land uses in the training set, with the test values validating this strong performance when new data were considered. A visual representation of the R2 is presented in Figure 1 in terms of scatter plots of the predicted versus observed 
-values. However, despite the strong correlation between the predicted and observed values, the MAPE values indicate that the model errors are in the order of 30% for the ‘General Residential’ model, and in the order of 40% for the ‘Low-Income Residential’. The relative size of the errors, also visible in Figure 1, suggests that the results should be interpreted conservatively. In view of the application of this model, at the early stages of a development, many unknowns would be introduced to the planning process. The relatively large model errors are considered acceptable given the other inherent uncertainties at this early stage of a development.
Close modal
-values. The test dataset values represent the model performance on data which were not used to train the model in its development. The R2 values were considered to be good, exceeding 0.7 for both land uses in the training set, with the test values validating this strong performance when new data were considered. A visual representation of the R2 is presented in Figure 1 in terms of scatter plots of the predicted versus observed -values. However, despite the strong correlation between the predicted and observed values, the MAPE values indicate that the model errors are in the order of 30% for the ‘General Residential’ model, and in the order of 40% for the ‘Low-Income Residential’. The relative size of the errors, also visible in Figure 1, suggests that the results should be interpreted conservatively. In view of the application of this model, at the early stages of a development, many unknowns would be introduced to the planning process. The relatively large model errors are considered acceptable given the other inherent uncertainties at this early stage of a development. Table 9
Training and test data R2 and MAPE for total pipeline length models
| Land use category | R2 | MAPE (%) | ||
|---|---|---|---|---|
| Training data | Test data | Training data | Test data | |
| General residential | 0.85 | 0.82 | 31.9 | 32.6 |
| Low-income residential | 0.74 | 0.98 | 40.7 | 35.3 |
| Land use category | R2 | MAPE (%) | ||
|---|---|---|---|---|
| Training data | Test data | Training data | Test data | |
| General residential | 0.85 | 0.82 | 31.9 | 32.6 |
| Low-income residential | 0.74 | 0.98 | 40.7 | 35.3 |
Figure 1
Model predicted versus observed values. (a) General residential training data. (b) General residential test data. (c) Low-income residential training data. (d) Low-income residential test data.
Figure 1
Model predicted versus observed values. (a) General residential training data. (b) General residential test data. (c) Low-income residential training data. (d) Low-income residential test data.