Table 5 Scaling of the overall test accuracy for prediction at lead day 1, lead day 3, and lead day 5 with sample size (N) for each of the methods (CNN4 and Log-Reg) for summer and winter.
Lead days | Sample Size | CNN4 Summer | CNN4 Winter | Log-Reg Summer | Log-Reg Winter |
|---|---|---|---|---|---|
Lead day 1 | N = 12000 | 92.1% ± 0.4% | 93.3% ± 0.2% | 65.3% ± 0.2% | 66.7% ± 0.3% |
N = 8000 | 88.3% ± 0.4% | 90.4% ± 0.4% | 60.3% ± 0.3% | 62.7% ± 0.5% | |
N = 4000 | 87.6% ± 0.4% | 86.5% ± 0.2% | 52.3% ± 0.4% | 54.6% ± 0.7% | |
Lead day 3 | N = 12000 | 83.4% ± 0.2% | 87.2% ± 0.2% | 59.7% ± 0.2% | 63.1% ± 0.3% |
N = 8000 | 81.3% ± 0.3% | 86.4% ± 0.0% | 52.3% ± 0.3% | 57.7% ± 0.5% | |
N = 4000 | 78.6% ± 0.5% | 83.4% ± 0.3% | 48.3% ± 0.5% | 50.6% ± 0.4% | |
Lead day 5 | N = 12000 | 76.4% ± 0.4% | 80.3% ± 0.2% | 55.4% ± 0.2% | 61.7% ± 0.3% |
N = 8000 | 74.2% ± 0.2% | 78.6% ± 0.5% | 51.3% ± 0.3% | 55.7% ± 0.5% | |
N = 4000 | 71.1% ± 0.5% | 74.4% ± 0.3% | 45.3% ± 0.3% | 49.6% ± 0.2% |
