Traffic alerts within 1 km of GTH and RCP (Page 34)

To assess the quality of the routes computed by the geocoder, we compared our geocoded routes with ground truth data provided by the DTP through their website (DTP 2019). The Delhi Traffic Police (DTP) reported a set of 50 known point locations identified as traffic congestion hotspots across the Delhi–NCR region which we refer to as ground truth hotspots (GTH). We then plotted these on a map along with our weighted complex network (see Figure 8). To obtain a numerical comparison, we counted the number of tweeted traffic alerts within a 1 km buffer of each GTH (see Figure 8). Our hypothesis is that if geocoding is accurate, areas in the immediate proximity (&lt; 1km) of GTH should have higher counts as compared to areas further away (1-5 km). To test this hypothesis, we randomly sampled 50 random control point locations (RCP) in the Delhi-NCR and computed the number of traffic alerts within a 1 km radius of those.<br>An independent-samples Welch’s t-test was used to compare the mean number of traffic alerts within a 1 km buffer of GTH and RCP sites. GTH sites (mean = 102, sd = 124, N = 50) recorded significantly more alerts than RCP sites (mean = 33, sd = 111, N = 50). The mean difference of 69 (SE = 24) was statistically significant, <i>t</i>(96) = 2.92, <i>p</i> = 0.0044, with a 95% confidence interval ranging from 22.00 to 115. The results indicate that traffic alert density is substantially higher (at least 22 more alerts on average at the 95% confidence level) around GTH sites compared to randomly selected control areas. This serves as statistically significant evidence to support the hypothesis that geocoding results accord spatially with ground truth data.

为评估地理编码器所生成路线的质量，我们将自身得到的地理编码路线与德里交通警察局（Delhi Traffic Police, DTP）通过其官方网站发布的真实基准数据（ground truth data，DTP 2019）进行了对比。德里交通警察局（Delhi Traffic Police, DTP）公布了德里-国家首都辖区（Delhi–NCR）范围内50处已知的交通拥堵热点点位，我们将其称为真实基准热点（ground truth hotspots, GTH）。随后我们将这些点位与我们构建的加权复杂网络一同绘制于地图之上（见图8）。为开展量化对比，我们统计了每一处真实基准热点周边1公里缓冲区范围内的推文式交通警报数量（见图8）。我们的研究假设为：若地理编码精度准确，则真实基准热点紧邻区域（<1公里）内的交通警报数量，应显著高于距离更远的区域（1至5公里）内的警报数量。为验证该假设，我们在德里-国家首都辖区内随机选取了50处对照点位（random control point, RCP），并统计了这些点位周边1公里半径范围内的交通警报数量。
我们采用独立样本Welch t检验，对比了真实基准热点与对照点位周边1公里缓冲区的平均交通警报数量。结果显示，真实基准热点点位的警报记录均值为102（标准差sd=124，样本量N=50），显著高于对照点位的均值33（标准差sd=111，样本量N=50）。二者平均差值为69（标准误SE=24），该差异具有统计学显著性：*t*(96) = 2.92，*p* = 0.0044，95%置信区间为22.00至115。研究结果表明，相较于随机选取的对照区域，真实基准热点周边的交通警报密度显著更高（在95%置信水平下，平均至少多出22条警报）。这一结果为“地理编码结果在空间上与真实基准数据相符”的假设提供了具备统计学显著性的支撑证据。