| import torch
|
| import json
|
| import numpy as np
|
| import matplotlib.pyplot as plt
|
| import seaborn as sns
|
| from sklearn.metrics import confusion_matrix, classification_report
|
| from torch_geometric.data import Data
|
| from gnn_aml import GAT, prepare_graph
|
| from graph_aml import detect_pattern
|
|
|
|
|
| print("π Loading Trained Model...")
|
| model = GAT(num_node_features=3, hidden_dim=16, output_dim=2)
|
| model.load_state_dict(torch.load("trained_model.pth"))
|
| model.eval()
|
|
|
|
|
| print("π₯ Loading New Test Transactions...")
|
| with open("test_transactions.json", "r") as f:
|
| test_transactions = json.load(f)
|
|
|
|
|
| print("π Preparing Test Graph Data...")
|
| test_graph, _ = prepare_graph()
|
|
|
|
|
| print("π§ Running Predictions...")
|
| with torch.no_grad():
|
| output = model(test_graph)
|
| probs = torch.softmax(output, dim=1)
|
| predictions = (probs[:, 1] > 0.75).long()
|
|
|
|
|
| test_results = []
|
| y_true = []
|
| y_pred = []
|
|
|
| for txn, prediction in zip(test_transactions, predictions):
|
| risk_score = txn["RiskScore"]
|
| true_label = 1 if txn["AML_Flag"] == 1 else 0
|
| predicted_label = prediction.item()
|
|
|
|
|
| y_true.append(true_label)
|
| y_pred.append(predicted_label)
|
|
|
| if risk_score < 0.5:
|
| predicted_pattern = "None"
|
| elif predicted_label == 1:
|
| predicted_pattern = detect_pattern(test_graph)
|
| else:
|
| predicted_pattern = "None"
|
|
|
| test_results.append({
|
| "TransactionID": txn["TransactionID"],
|
| "TrueLabel": true_label,
|
| "PredictedLabel": predicted_label,
|
| "PredictedPattern": predicted_pattern,
|
| "RiskScore": risk_score
|
| })
|
|
|
|
|
| with open("new_test_results_v2.json", "w") as f:
|
| json.dump(test_results, f, indent=4)
|
|
|
|
|
| print("\nπ **Final Test Results:**")
|
| cm = confusion_matrix(y_true, y_pred)
|
| report = classification_report(y_true, y_pred, target_names=[
|
| "Normal", "AML"], digits=4)
|
|
|
| print("\nπ’ **Confusion Matrix:**\n", cm)
|
| print("\nπ **Classification Report:**\n", report)
|
|
|
|
|
| plt.figure(figsize=(6, 5))
|
| sns.heatmap(cm, annot=True, fmt="d", cmap="Blues", xticklabels=[
|
| "Normal", "AML"], yticklabels=["Normal", "AML"])
|
| plt.xlabel("Predicted")
|
| plt.ylabel("Actual")
|
| plt.title("Confusion Matrix")
|
| plt.show()
|
|
|
|
|
| labels, counts = np.unique(y_pred, return_counts=True)
|
| plt.figure(figsize=(6, 5))
|
| plt.bar(["Normal", "AML"], counts, color=["green", "red"])
|
| plt.xlabel("Transaction Classification")
|
| plt.ylabel("Number of Transactions")
|
| plt.title("AML vs. Normal Transactions Detected")
|
| plt.show()
|
|
|
| print("β
Accuracy analysis complete! Check charts & logs.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|