Smart_Container_Risk_Prediction

SmartContainer Risk Engine

An AI/ML-powered shipment risk prediction system designed to identify suspicious or high-risk cargo containers using historical customs and shipment data.

The system combines:

• Machine Learning Classification
• Anomaly Detection
• Explainable AI
• Risk Analytics
• Real-Time Shipment Monitoring

---

Problem Statement

Traditional container inspection systems rely heavily on manual verification and static rules, making it difficult to detect suspicious shipment activities efficiently.

This project aims to build an intelligent risk engine capable of:

• Detecting fraudulent declarations
• Identifying suspicious trade routes
• Flagging risky importers/exporters
• Detecting abnormal shipment behavior
• Supporting customs and logistics security operations

---

Features

Advanced Feature Engineering

• Weight difference analysis
• Weight deviation percentage
• Value per KG calculation
• Trade route generation
• Night-time declaration detection
• High dwell-time risk detection

---

Dynamic Risk Scoring

Risk scores generated using:

• Importer history
• Exporter history
• Shipping line history
• HS Code historical risk
• Port risk analysis
• Trade route risk patterns

---

Machine Learning Models

Random Forest Classifier

Used for:

• Shipment risk prediction
• Risk probability estimation

Isolation Forest

Used for:

• Outlier detection
• Unknown anomaly identification

---

Explainable AI

Generates human-readable risk explanations such as:

• Large weight difference detected
• Shipment stayed unusually long at port
• Importer has previous risky shipment history
• Trade route historically high risk

---

Real-Time Risk Prediction

Outputs:

• Risk Probability
• Risk Level
• Final Risk Decision
• Risk Explanation

---

Tech Stack

Technology	Purpose
Python	Core Programming
Pandas	Data Processing
NumPy	Numerical Operations
Scikit-learn	Machine Learning
Random Forest	Classification
Isolation Forest	Anomaly Detection

---

Workflow

1️ Load Data

• Historical shipment data
• Real-time shipment data

---

2️ Feature Engineering

Creates intelligent features like:

• Weight_Diff
• Weight_Diff_Percent
• Value_per_KG
• Trade_Route
• Night_Declaration
• High_Dwell_Risk

---

3️ Safe Categorical Encoding

Encodes categorical variables using mappings built only from historical data to prevent data leakage.

---

4️ Aggregated Risk Features

Builds:

• Shipment frequency metrics
• Importer/exporter behavioral patterns
• Historical risk scores

---

5️ Model Validation

Evaluation metrics used:

• Macro F1 Score
• Weighted F1 Score
• Recall Score
• Confusion Matrix
• Classification Report

---

6️ Production Training

Final models trained using the complete historical dataset.

---

7️ Real-Time Prediction

Predicts:

• Risk Probability
• Risk Level
• Final Risk Status

---

8️ Explainable Risk Engine

Provides reasons behind risk predictions.

---

Contributed by: [Aarju Patel]