chore: Add model selection section to the notebook

notebook_guide.ipynb

@@ -95700,7 +95700,31 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## 5. Model Selection\n"
+    "## 5. Model Comparison and Selection\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "The following table summarizes the performance of all models trained in this project.\n",
+    "\n",
+    "| Model               | Modality     | Accuracy | Macro Avg F1-Score | Weighted Avg F1-Score |\n",
+    "| :------------------ | :----------- | :------- | :----------------- | :-------------------- |\n",
+    "| Random Forest       | Text         | 0.90     | 0.83               | 0.90                  |\n",
+    "| Logistic Regression | Text         | 0.90     | 0.84               | 0.90                  |\n",
+    "| Random Forest       | Image        | 0.80     | 0.70               | 0.79                  |\n",
+    "| Random Forest       | Combined     | 0.89     | 0.79               | 0.89                  |\n",
+    "| Logistic Regression | Combined     | 0.89     | 0.83               | 0.89                  |\n",
+    "| **MLP**             | **Image**    | **0.84** | **0.77**           | **0.84**              |\n",
+    "| **MLP**             | **Text**     | **0.92** | **0.87**           | **0.92**              |\n",
+    "| **MLP**             | **Combined** | **0.92** | **0.85**           | **0.92**              |\n",
+    "\n",
+    "Based on the overall results, the MLP models consistently outperformed their classical machine learning counterparts. This is a common outcome when dealing with complex, high-dimensional features like the embeddings we are using. While traditional models like Random Forest and Logistic Regression are powerful, neural networks like the MLP are specifically designed to learn intricate, non-linear relationships within the data, which often leads to a performance edge on tasks like this.\n",
+    "\n",
+    "Both the Text MLP and Combined MLP models achieved the highest accuracy and weighted F1-score, confirming their superior ability to classify the products.\n",
+    "\n",
+    "For this project's final implementation, we will proceed with the **Multilayer Perceptron (MLP) models**. Their performance, particularly on the combined text and image data, makes them the ideal choice for a robust classification system that can leverage all available product information. These models are now ready to be used in a future application to perform real-time predictions.\n"
    ]
   }
  ],