predicted_summary_stats_ecoli

The published dataset shows the summary statistics of predicted <em>E.coli </em>concentrations at the Seine river in Paris at Pont d'Iena. The data is provided on a log₁₀scale and shows the geometric mean and the predicted upper percentiles, which can be compared to the thresholds provided by the European Bathing Water Directive. Model predictions have a root mean squared error of 0.25 and a coverage rate of 97%, 95%, and 98% for the ratio of test data points falling below the 95th, the 90th percentile, and inside the 95% prediction interval. However, model predictions are very uncertain and are not able to accurately predict variations of measured <em>E.coli</em> concentrations. Therefore, the model is not able to discriminate between periods of good and impaired water quality and always predicts "poor" water quality. The used features were not suitable to accurately predict the observed concentrations. Further model improvements are considered necessary.本公开数据集展示了巴黎塞纳河耶拿桥（Pont d'Iena）点位的预测大肠杆菌（E. coli）浓度统计摘要。数据以10为底的对数尺度（log₁₀ scale）呈现，包含几何均值与预测上百分位数，可与《欧盟沐浴水指令》（European Bathing Water Directive）规定的阈值进行比对。模型预测的均方根误差为0.25；针对低于95百分位数、低于90百分位数，以及处于95%预测区间内的测试数据点占比，其覆盖率分别为97%、95%与98%。然而，该模型的预测不确定性极强，无法准确复现实测大肠杆菌浓度的变化规律，亦无法区分水质良好与受损的时段，始终输出“劣质”水质的预测结果。本次研究所采用的特征变量不足以精准预测观测得到的浓度值，因此亟需对模型开展进一步优化完善。