Delays_Are_Not_Random/data_sources_and_features.txt at main · LozanoLsa/Delays_Are_Not_Random · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
PROJECT: 01_Logistic_Regression_SUP
TITLE: Data Sources & Feature Engineering

--------------------------------------------------
OVERVIEW:
This dataset contains 1,847 shipment records from a road freight logistics operation across three route categories and two departure shifts. The ML task is binary classification to predict whether a shipment will arrive delayed (1) or on time (0). Class balance: 68.0% delayed — reflecting a structural operational problem, not a random event. Source: ERP/TMS operational data covering distance, load, facility timing, and operator assignment.

--------------------------------------------------
REAL-WORLD DATA SOURCES:

Feature               | Source System              | Instrument
----------------------|----------------------------|--------------------------
distance_km           | Route Planning / TMS       | Route planning system records
num_stops             | Route Planning / TMS       | Delivery stop schedule
route_type            | Route Planning / TMS       | Route classification (urban / mixed / long_haul)
weight_kg             | ERP / Shipment Master      | Cargo weight at shipment entry
priority              | ERP / Shipment Master      | Priority flag (1 = urgent, 0 = standard)
loading_time_min      | Warehouse / Dock Execution Log | Dock time-stamp system
dock_wait_time_min    | Warehouse / Dock Execution Log | Pre-loading wait time log
departure_shift       | Shipment Release Timestamps | TMS departure event log
operator_experience_yrs | Workforce / Operator Assignment Records | HR system / operator profile
delayed               | Historical Delivery Compliance Records | TMS delivery confirmation log

--------------------------------------------------
VARIABLE DESCRIPTIONS:

distance_km            - Route distance in kilometers; longer routes increase transit exposure.
weight_kg              - Cargo weight in kilograms; affects handling and transportation conditions.
num_stops              - Number of intermediate delivery stops; more stops increase delay risk.
priority               - Binary flag: 1 = urgent shipment, 0 = standard.
loading_time_min       - Time spent loading at origin dock (min); longer loading creates schedule deviation.
dock_wait_time_min     - Pre-loading wait at dock (min); represents congestion and operational waiting.
operator_experience_yrs - Driver/operator experience (years); reduces variability and improves execution consistency.
route_type             - Categorical: urban / mixed / long_haul. Urban routes have highest delay rate (74.5%) vs 57.7% for long-haul.
departure_shift        - Categorical: day / night. Night departures delay at 72.6% vs 64.9% for day shifts.
delayed                - Binary target: 1 = late, 0 = on time. Class balance: 68.0% delayed.

--------------------------------------------------
FEATURE ENGINEERING:

StandardScaler applied to all numerical features. Label Encoding applied to categorical features (route_type, departure_shift). All preprocessing inside a sklearn Pipeline. Train/Test split: 80/20, random_state=42.

--------------------------------------------------
BUSINESS CONTEXT:

Used by logistics and dispatch operations teams to score shipment delay risk before a shipment leaves the dock. With 68% of shipments arriving delayed, the problem is structural and requires predictive intervention rather than reactive tracking.

--------------------------------------------------
GOAL:

Predict delay probability before departure to enable risk-based prioritization of the shipment queue, improve OTIF (On Time In Full), and support better dispatch decisions. Model achieved Accuracy 69.7%, Recall (Delay) 91.3%, AUC-ROC 0.662.