diff --git a/readme.md b/readme.md
index 2586ad8..c34791e 100644
--- a/readme.md
+++ b/readme.md
@@ -244,6 +244,7 @@ for epoch in range(num_epochs + 1):
     if epoch % 10 == 0:
         # 모델 저장
         torch.save(cnn_wdi.state_dict(), 'CNN_WDI_' + str(epoch) + 'epoch.pth')
+```
 
 -----
 
@@ -251,6 +252,219 @@ for epoch in range(num_epochs + 1):
 
 * 모델의 성능지표(Precision, Recall, Accuracy, F1-Score)를 혼동행렬(Confusion Metrix)로 구현한다.
 
+# Confusion Metrix
+
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+import seaborn as sns
+from sklearn.metrics import classification_report, confusion_matrix
+import pandas as pd
+
+def plot_metrics(title, class_names, precisions, recalls, f1_scores, acc):
+    num_classes = len(class_names)
+    index = np.arange(num_classes)
+    bar_width = 0.2
+
+    plt.figure(figsize=(15, 7))
+    plt.bar(index, precisions, bar_width, label='Precision')
+    plt.bar(index + bar_width, recalls, bar_width, label='Recall')
+    plt.bar(index + 2 * bar_width, f1_scores, bar_width, label='F1-score')
+    plt.axhline(y=acc, color='r', linestyle='--', label='Accuracy')
+
+    plt.xlabel('Class')
+    plt.ylabel('Scores')
+    plt.title(title + ': Precision, Recall, F1-score, and Accuracy per Class')
+    plt.xticks(index + bar_width, class_names)
+    plt.legend(loc='upper right')
+    plt.show()
+
+def predict_and_plot_metrics(title, model, dataloader, criterion, device):
+    model.eval()
+    running_loss = 0.0
+    running_corrects = 0
+
+    all_preds = []
+    all_labels = []
+    class_names = ['Center', 'Donut', 'Edge-Loc', 'Edge-Ring', 'Loc', 'Near-full', 'none', 'Random', 'Scratch']
+
+    with torch.no_grad():
+        for inputs, labels in dataloader:
+            inputs = inputs.to(device)
+            labels = labels.to(device)
+
+            outputs = model(inputs)
+            _, preds = torch.max(outputs, 1)
+            loss = criterion(outputs, labels)
+
+            running_loss += loss.item() * inputs.size(0)
+            running_corrects += torch.sum(preds == labels.data)
+
+            all_preds.extend(preds.cpu().numpy())
+            all_labels.extend(labels.cpu().numpy())
+
+        epoch_loss = running_loss / len(dataloader.dataset)
+        epoch_acc = running_corrects.double() / len(dataloader.dataset)
+
+
+    # Calculate classification report
+    report = classification_report(all_labels, all_preds, target_names=class_names, output_dict=True)
+
+    # Calculate confusion matrix
+    cm = confusion_matrix(all_labels, all_preds)
+
+    # Calculate precision, recall, and f1-score per class
+    precisions = [report[c]['precision'] for c in class_names]
+    recalls = [report[c]['recall'] for c in class_names]
+    f1_scores = [report[c]['f1-score'] for c in class_names]
+    print('p: ' + str(precisions))
+    print('r: ' + str(recalls))
+    print('f: ' + str(f1_scores))
+
+    # Plot confusion matrix with normalized values (percentage)
+    cm_normalized = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
+    plt.figure(figsize=(12, 12))
+    sns.heatmap(cm_normalized, annot=True, fmt='.2%', cmap='Blues', xticklabels=class_names, yticklabels=class_names)
+    plt.xlabel('Predicted Label')
+    plt.ylabel('True Label')
+    plt.title('Normalized Confusion Matrix: ' + title)
+    plt.show()
+
+    # Plot precision, recall, f1-score, and accuracy per class
+    plot_metrics(title, class_names, precisions, recalls, f1_scores, epoch_acc.item())
+
+    return epoch_loss, epoch_acc, report
+
+```
+
+# Evaluate
+
+
+```python
+import os
+import re
+
+test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=112, shuffle=False, num_workers=4)
+
+dir = '.'
+models = [file for file in os.listdir(dir) if file.endswith(('.pth'))]
+
+def extract_number(filename):
+    return int(re.search(r'\d+', filename).group(0))
+
+sorted_models = sorted(models, key=extract_number)
+
+for model in sorted_models:
+    model_path = os.path.join(dir, model)
+
+    # Load the saved model weights
+    cnn_wdi.load_state_dict(torch.load(model_path))
+
+    # Call the predict_and_plot_metrics function with the appropriate arguments
+    epoch_loss, epoch_acc, report = predict_and_plot_metrics(model, cnn_wdi, test_loader, criterion, device)
+    # print(f'Model: {model} Test Loss: {test_loss:.4f} Acc: {test_acc:.4f}')
+```
+
+# Loss Graph
+
+
+```python
+import matplotlib.pyplot as plt
+
+# 파일에서 데이터를 읽어들입니다.
+with open('output.txt', 'r') as file:
+    lines = file.readlines()[1:]  # 첫 번째 줄은 헤더이므로 건너뜁니다.
+
+# 데이터를 분석하여 리스트에 저장합니다.
+epochs = []
+train_losses = []
+train_accuracies = []
+val_losses = []
+val_accuracies = []
+
+for line in lines:
+    if line == '\n':
+        continue
+    # epoch, train_loss, train_acc, val_loss, val_acc = line.strip().split(', \t')
+    epoch, train_loss, train_acc, val_loss, val_acc = re.split(r'[,\s\t]+', line.strip())
+    epochs.append(int(epoch.split('/')[0]))
+    train_losses.append(float(train_loss))
+    train_accuracies.append(float(train_acc))
+    val_losses.append(float(val_loss))
+    val_accuracies.append(float(val_acc))
+
+# 선 그래프를 그립니다.
+plt.figure(figsize=(10, 5))
+
+plt.plot(epochs, train_losses, label='Train Loss')
+plt.plot(epochs, train_accuracies, label='Train Acc')
+plt.plot(epochs, val_losses, label='Val Loss')
+plt.plot(epochs, val_accuracies, label='Val Acc')
+
+plt.xlabel('Epochs')
+plt.ylabel('Values')
+plt.title('Training and Validation Loss and Accuracy')
+plt.legend()
+plt.show()
+
+```
+
+# Print Selecting Test Model Result
+
+
+```python
+def output(model, dataloader, criterion, device):
+    model.eval()
+    running_loss = 0.0
+    running_corrects = 0
+
+    all_preds = []
+    all_labels = []
+    class_names = ['Center', 'Donut', 'Edge-Loc', 'Edge-Ring', 'Loc', 'Near-full', 'none', 'Random', 'Scratch']
+
+    with torch.no_grad():
+        for inputs, labels in dataloader:
+            inputs = inputs.to(device)
+            labels = labels.to(device)
+
+            outputs = model(inputs)
+            _, preds = torch.max(outputs, 1)
+            loss = criterion(outputs, labels)
+
+            running_loss += loss.item() * inputs.size(0)
+            running_corrects += torch.sum(preds == labels.data)
+
+            all_preds.extend(preds.cpu().numpy())
+            all_labels.extend(labels.cpu().numpy())
+
+        epoch_loss = running_loss / len(dataloader.dataset)
+        epoch_acc = running_corrects.double() / len(dataloader.dataset)
+
+
+    # Calculate classification report
+    report = classification_report(all_labels, all_preds, target_names=class_names, output_dict=True)
+
+    # Calculate precision, recall, and f1-score per class
+    precisions = [report[c]['precision'] for c in class_names]
+    recalls = [report[c]['recall'] for c in class_names]
+    f1_scores = [report[c]['f1-score'] for c in class_names]
+    accuracy = report['accuracy']
+
+    precs = sum(precisions) / len(precisions)
+    recs = sum(recalls) / len(recalls)
+    f1s = sum(f1_scores) / len(f1_scores)
+    print('precisions: ' + str(precs))
+    print('recalls: ' + str(recs))
+    print('f1_scores: ' + str(f1s))
+    print('accuracy ' + str(accuracy))
+
+
+selected_model = 'CNN_WDI_20epoch.pth'
+cnn_wdi.load_state_dict(torch.load(selected_model))
+output(cnn_wdi, test_loader, criterion, device)
+```
+
 -----
 
 ### 테스트 결과