UK Road Collision Severity Analysis 2024

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
import seaborn as sns
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import warnings
warnings.filterwarnings('ignore')

# Set visual defaults
matplotlib.rcParams['figure.dpi'] = 120
matplotlib.rcParams['figure.facecolor'] = 'white'
sns.set_style('whitegrid')
sns.set_palette('viridis')

print("Libraries loaded. Reading dataset...")

Libraries loaded. Reading dataset...

# ============================================================
# 1. LOAD AND PROFILE
# ============================================================
df = pd.read_csv('dft-road-casualty-statistics-collision-2024.csv',
                 low_memory=False)

print(f"Shape: {df.shape}")
print(f"\n=== COLUMN DTYPES ===")
print(df.dtypes.to_string())

print(f"\n=== FIRST 3 ROWS (key columns) ===")
key_cols = ['collision_index','collision_severity','date','time','speed_limit',
            'road_type','junction_detail','light_conditions','weather_conditions',
            'road_surface_conditions','urban_or_rural_area','latitude','longitude',
            'number_of_vehicles','number_of_casualties']
print(df[key_cols].head(3).to_string())

print(f"\n=== UNIQUE VALUES PER COLUMN ===")
…

Shape: (100927, 44)

=== COLUMN DTYPES ===
collision_index                                      object
collision_year                                        int64
collision_ref_no                                     object
location_easting_osgr                                 int64
location_northing_osgr                                int64
longitude                                           float64
latitude                                            float64
police_force…

# ============================================================
# 2. CLEAN AND DECODE THE DATA
# ============================================================

df = pd.read_csv('dft-road-casualty-statistics-collision-2024.csv', low_memory=False)

# --- Severity (target) ---
severity_map = {1: 'Fatal', 2: 'Serious', 3: 'Slight'}
df['severity_label'] = df['collision_severity'].map(severity_map)
# Binary target: 1 = Fatal or Serious, 0 = Slight
df['is_serious_or_fatal'] = df['collision_severity'].isin([1, 2]).astype(int)

# --- Day of week (1=Sunday..7=Saturday per DfT STATS20) ---
dow_map = {1:'Sunday',2:'Monday',3:'Tuesday',4:'Wednesday',
           5:'Thursday',6:'Friday',7:'Saturday'}
df['day_name'] = df['day_of_week'].map(dow_map)

# --- Datetime (date format is DD/MM/YYYY) ---
…

=== MISSINGNESS SUMMARY ===
                                 missing   pct
light_label                            6  0.01
local_authority_highway_current        3  0.00

=== SEVERITY DISTRIBUTION ===
severity_label
Slight     75858
Serious    23567
Fatal       1502
Name: count, dtype: int64

Binary target share: 0.248 (25,069 of 100,927 crashes)

=== DATE RANGE ===
From 2024-01-01 00:00:00 to 2024-12-31 00:00:00
Rows with invalid date: 0

=== DECODED KEY COLUMNS (sample) ===
   collision_severit…

# ============================================================
# 3. EDA — Time, Speed Limits, Weather, Lighting, Road Surface, Junctions
# ============================================================

fig, axes = plt.subplots(2, 3, figsize=(18, 11))
fig.suptitle('Collision Severity by Key Factors — UK 2024', fontsize=16, fontweight='bold', y=1.01)

# --- 1. Hour of day ---
sev_hour = df.groupby(['hour', 'severity_label']).size().unstack(fill_value=0)
sev_hour_pct = sev_hour.div(sev_hour.sum(axis=1), axis=0)
ax = axes[0,0]
for sev, color in zip(['Slight','Serious','Fatal'], ['#2ecc71','#f39c12','#e74c3c']):
    if sev in sev_hour_pct:
        ax.plot(sev_hour_pct.index, sev_hour_pct[sev], label=sev, color=color, lw=2)
ax.set_xlabel('Hour of day'); ax.set_ylabel('Proportion within hour')
ax.set_title('Severity by Hour of Day', fontsize=12)
ax.legend(fontsize=9); ax.set_xticks(range(0,24,2)); ax.grid(alpha=0.3)

…

================================================================================
SERIOUS/FATAL RATE BY FACTOR (top categories)
================================================================================

--- Speed Limit ---
             N_crashes  serious_fatal_rate  N_serious_fatal
speed_limit                                                
 30              50257               0.231            11629
 20              18881               0.209             3953
 60              12709…

# ============================================================
# 4. HOTSPOT MAP — Geographic concentration of serious/fatal crashes
# ============================================================

import plotly.express as px

# Filter to crashes with valid lat/lon
map_df = df.dropna(subset=['latitude', 'longitude']).copy()
# Clip to GB mainland roughly
map_df = map_df[(map_df['latitude'] > 49.8) & (map_df['latitude'] < 58.7) &
                (map_df['longitude'] > -8.5) & (map_df['longitude'] < 2.5)]

# Separate serious/fatal from slight for layering
sf = map_df[map_df['is_serious_or_fatal'] == 1]
sl = map_df[map_df['is_serious_or_fatal'] == 0]

# Sample for performance (Plotly can struggle with 100K points)
sf_sample = sf.sample(min(8000, len(sf)), random_state=42)
…

Saved to outputs (downloadable from the file manager):
  • collision_hexbin_density.png (364806 bytes)
  • top_police_forces_sf.png (74435 bytes)
Not persisted:
  • collision_hotspot_map.html — could not read from sandbox

# ============================================================
# 5. PREDICTIVE MODEL — Factors associated with serious/fatal crashes
# ============================================================

from sklearn.model_selection import train_test_split, cross_val_score, StratifiedKFold
from sklearn.ensemble import GradientBoostingClassifier, RandomForestClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import LabelEncoder, StandardScaler
from sklearn.metrics import (roc_auc_score, classification_report, confusion_matrix,
                             RocCurveDisplay, PrecisionRecallDisplay,
                             average_precision_score)
from sklearn.inspection import permutation_importance
import warnings
warnings.filterwarnings('ignore')

# ---- 5a. Prepare feature matrix ----
# Drop invalid speed limits
model_df = df[df['speed_limit'] > 0].copy()
…

Feature matrix shape: (100924, 14)
Target: 25,068 serious/fatal out of 100,924 (0.248)

Features used: ['speed_limit', 'road_type', 'junction_detail', 'junction_control', 'light_conditions', 'weather_conditions', 'road_surface_conditions', 'urban_or_rural_area', 'hour', 'day_of_week', 'month', 'number_of_vehicles', 'special_conditions_at_site', 'carriageway_hazards']

Train: 80,739  |  Test: 20,185
Train SF rate: 0.248  |  Test SF rate: 0.248

====================================================…