Analytics

Statistical analysis · Z-score normalization · Correlation & distribution research

Mean (μ)
643
detections / behavior
Median
519
center of distribution
Std Dev (σ)
495
spread across behaviors
IQR
441
Q3 − Q1 range
Top z-score
+2.43σ
RB-01
HIGH severity
67%
of all detections
Behavior Z-Score Analysis
Standard deviations from mean (μ = 643, σ = 495) · each bar colored by behavior class
z > 1σ0–1σbelow avg
↳ Insight:RB-01 (Helmet Non-Compliance) is the strongest outlier at 2.43σ above mean — nearly 2.4× the standard deviation, indicating disproportionate prevalence compared to other behaviors. 2 behavior(s) exceed the +1σ threshold, confirming a right-skewed distribution.
Pareto Analysis (80/20 Rule)
Count per class (bars) + cumulative % (line) · dashed = 80% threshold
↳ Insight:Only 6 of 10 behaviors account for ≥80% of all detections (84.5%). Targeted interventions on RB-01, RB-02, RB-03, RB-04, RB-05, RB-06 alone could address the vast majority of school-zone violations.
30-Day Distribution (Box Plot)
Whiskers = min/max · box = IQR · line = median · ◆ = mean
1129476482RB-01RB-02RB-03RB-04RB-05RB-06Whisker (min/max)IQR (Q1–Q3)MedianMean ◆
↳ Insight:RB-01 has the widest IQR (21 detections/day), indicating high daily variability — likely tied to enforcement presence. Behaviors with narrow IQRs (RB-05, RB-06) show more predictable patterns, suitable for automated alert thresholds.
30-Day Detection Trend
Daily counts per behavior class · March – April 2026
↳ Insight:Multi-behavior trends reveal synchronised spikes on school days (Mon–Fri) with dips on weekends, confirming school-zone correlation. RB-01 consistently leads all classes with a daily mean of 61.4 detections. Correlated oscillations across behaviors suggest shared causal factors such as peak commuter traffic volume.
Inter-Behavior Correlation Matrix
Pearson r from 30-day daily series · positive · negative · diagonal = self (r = 1.0)
RB-01RB-02RB-03RB-04RB-05RB-06RB-07RB-08RB-09RB-10RB-011.00.6-0.6-0.60.8-0.5-0.80.4-0.7RB-020.61.00.4-0.5-0.70.9-0.3RB-03-0.61.00.4-0.70.70.60.6RB-04-0.60.41.00.50.80.60.9RB-050.80.4-0.71.00.4-0.60.3RB-06-0.50.50.41.00.3-0.50.4RB-07-0.50.70.81.00.50.8RB-08-0.8-0.70.60.6-0.60.30.51.0-0.50.6RB-090.40.90.3-0.5-0.51.0RB-10-0.7-0.30.60.90.40.80.61.0
↳ Insight:Strongest pair: RB-04 ↔ RB-10 (r = 0.92). High positive correlations suggest behavioral co-occurrence — riders who commit one violation are statistically more likely to commit another simultaneously, supporting a "risk profile clustering" hypothesis.
95% Confidence Intervals
Daily mean detection rate ± 95% CI (t-dist, df = 29) per behavior
↳ Insight:RB-01 shows the widest 95% CI (±5.01/day), indicating high day-to-day variability and lower statistical confidence. Narrower CIs on lower-frequency behaviors reflect more stable detection patterns, improving the reliability of trend-based predictions.
Risk Impact Matrix
Detection frequency index (x) vs. risk impact score (y) · bubble size = relative volume
↳ Insight:RB-01 and RB-02 occupy the critical quadrant (high frequency + high risk) — requiring immediate, scaled intervention. RB-07 (Pedestrian Conflict) sits in the high-risk/lower-frequency quadrant, demanding pre-emptive structural solutions despite lower current detection volume.
Normal Distribution of Detection Counts
Theoretical N(μ=643, σ=495) · each behavior marked at its observed count
RB-01RB-02RB-03RB-04RB-05RB-06RB-07RB-08RB-09RB-10-2σ-1σμ+1σ+2σ
RB-01
RB-02
RB-03
RB-04
RB-05
RB-06
RB-07
RB-08
RB-09
RB-10
↳ Insight:The distribution is strongly right-skewed — RB-01 and RB-02 fall 2.43σ and 1.13σ above the mean, far outside the expected normal range. 7 behaviors fall below average, confirming that a minority of violation types drive the majority of risk.
School Comparison
Submissions by behavior class per school (stacked)
↳ Insight:Lê Quý Đôn and Nguyễn Thị Minh Khai show the highest absolute detection volume. The consistent helmet non-compliance share (≈38%) across all schools suggests a systemic cultural factor rather than site-specific infrastructure, supporting a city-wide behavioral intervention strategy.
Time-of-Day Distribution
Detection frequency by hour · school day average
↳ Insight:Two sharp peaks emerge at 07:30 (234 detections) and 17:30 (234 detections), precisely coinciding with school arrival and dismissal windows. Off-peak periods drop 73% below peak, confirming that enforcement and AI-monitoring resources should be concentrated within ±30 minutes of school bell times.
Behavior Statistical Summary
Sorted by z-score · normalization metrics for comparative research analysis
CodeBehaviorSeverityTotalMean / DayZ-ScorePercentile% Share
RB-01Helmet Non-ComplianceHIGH1.84261.4+2.43σ
100th
28.7%
RB-02Phone Use While RidingHIGH1.20340.1+1.13σ
89th
18.7%
RB-03Wrong-Way RidingHIGH73124.4+0.18σ
78th
11.4%
RB-04Lane ViolationMODERATE61420.5-0.06σ
67th
9.6%
RB-05Red Light RunningMODERATE54818.3-0.19σ
56th
8.5%
RB-06Double/Triple RidingMODERATE49016.3-0.31σ
44th
7.6%
RB-07Pedestrian ConflictHIGH37812.6-0.53σ
33th
5.9%
RB-08Abrupt Stop in FlowMODERATE2909.7-0.71σ
22th
4.5%
RB-09OverloadingMODERATE1876.2-0.92σ
11th
2.9%
RB-10Speeding in School ZoneHIGH1424.7-1.01σ
0th
2.2%