1. CNN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Baseline CNN and creation of BSD\n",
    "\n",
    "### This file has the results of the baseline CNN and the creation of BSD dataset. The steps performed are:\n",
    "\n",
    "1. This dataset consists of 2302 training and 998 test samples. \n",
    "2. The experimental dataset is randomly split into train and test sets with a split ratio of 70:30.\n",
    "3. The baseline CNN model has four convolutional layers with 15 filters each of size 3*3. \n",
    "4. Adam's technique is used as the optimizer. Binary cross-entropy is used as the loss function. \n",
    "5. This CNN is trained for 150 epochs. \n",
    "6. Softmax is used as the final activation function which outputs predicted probabilities. \n",
    "7. The BSD is created which contains all the samples whose confidence factor is less than a certain threshold"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import scipy.misc\n",
    "import random\n",
    "import imageio\n",
    "from keras.utils import to_categorical\n",
    "import tensorflow as tf\n",
    "from tensorflow.python.tools import freeze_graph\n",
    "from tensorflow.python.tools import optimize_for_inference_lib\n",
    "from keras.applications.inception_v3 import InceptionV3\n",
    "\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.decomposition import PCA\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sklearn.utils import shuffle\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "from keras.utils.vis_utils import plot_model\n",
    "from keras import backend as K\n",
    "import keras.optimizers as optimizers\n",
    "from keras.utils import np_utils\n",
    "from keras.models import Sequential\n",
    "from keras.layers.core import Dense, Dropout, Activation, Flatten\n",
    "from keras.models import load_model\n",
    "from keras.layers import Lambda, Conv2D, MaxPooling2D, Dropout, Dense, Flatten, Activation, BatchNormalization\n",
    "from keras.optimizers import SGD,RMSprop,Adam"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The malignant and benign data is loaded to np arrays. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "xs = []\n",
    "ys = []\n",
    "def Load_data_malignant():\n",
    "    path =\"Dataset_Final/malignant\"\n",
    "    x_out = []\n",
    "    y_out = []\n",
    "    for i in range(1, 1500):\n",
    "        img = imageio.imread(path +'/' + str(i) + '.jpg')\n",
    "        lab = 1 \n",
    "        x_out.append(img)\n",
    "        y_out.append(lab)\n",
    "    return x_out, y_out\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def Load_data_benign():\n",
    "    path =\"Dataset_Final/benign\"\n",
    "    x_out = []\n",
    "    y_out = []\n",
    "    for i in range(1, 1800):\n",
    "        img = imageio.imread(path +'/' + str(i) + '.jpg')\n",
    "        lab = 0 \n",
    "        x_out.append(img)\n",
    "        y_out.append(lab)\n",
    "    return x_out, y_out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The benign and malignant arrays are assigned to x and y arrays"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_0, y_0 = Load_data_benign()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_1, y_1 = Load_data_malignant()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_0 = np.array(x_0)\n",
    "y_0 = np.array(y_0)\n",
    "\n",
    "x_1 = np.array(x_1)\n",
    "y_1 = np.array(y_1)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The arrays are concatenated to form x and y arrays (features, outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_classes = 2\n",
    "x=np.concatenate((x_0, x_1), axis=0)\n",
    "y=np.concatenate((y_0, y_1), axis=0)\n",
    "\n",
    "y = np_utils.to_categorical(y, num_classes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The data is split as train and test with 70:30 split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train, x_Test, y_train, y_Test = train_test_split(x, y, test_size=0.3, random_state=5,stratify=y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The CNN model is defined below"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "nClasses = 2\n",
    "\n",
    "def createModel():\n",
    "    model = Sequential()\n",
    "    \n",
    "    model.add(Conv2D(15, (3, 3), padding='valid', activation='relu', input_shape=(224,224,3)))\n",
    "    model.add(BatchNormalization())\n",
    "      \n",
    "    model.add(Conv2D(15, (3, 3), padding='same',activation='relu'))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    model.add(Dropout(0.3))\n",
    " \n",
    "    model.add(Conv2D(15, (3, 3), padding='same',activation='relu'))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    model.add(Dropout(0.3))\n",
    "    \n",
    "    model.add(Conv2D(15, (3, 3), padding='valid', activation='relu'))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(100, activation='relu'))\n",
    "    model.add(Dropout(0.3))\n",
    "    model.add(Dense(nClasses, activation='softmax'))\n",
    "    model.summary()\n",
    "     \n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Optimizer is chosen as Adam, binary cross entropy is the loss function, with epochs as 150"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " conv2d (Conv2D)             (None, 222, 222, 15)      420       \n",
      "                                                                 \n",
      " batch_normalization (BatchN  (None, 222, 222, 15)     60        \n",
      " ormalization)                                                   \n",
      "                                                                 \n",
      " conv2d_1 (Conv2D)           (None, 222, 222, 15)      2040      \n",
      "                                                                 \n",
      " batch_normalization_1 (Batc  (None, 222, 222, 15)     60        \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " max_pooling2d (MaxPooling2D  (None, 111, 111, 15)     0         \n",
      " )                                                               \n",
      "                                                                 \n",
      " dropout (Dropout)           (None, 111, 111, 15)      0         \n",
      "                                                                 \n",
      " conv2d_2 (Conv2D)           (None, 111, 111, 15)      2040      \n",
      "                                                                 \n",
      " batch_normalization_2 (Batc  (None, 111, 111, 15)     60        \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " max_pooling2d_1 (MaxPooling  (None, 55, 55, 15)       0         \n",
      " 2D)                                                             \n",
      "                                                                 \n",
      " dropout_1 (Dropout)         (None, 55, 55, 15)        0         \n",
      "                                                                 \n",
      " conv2d_3 (Conv2D)           (None, 53, 53, 15)        2040      \n",
      "                                                                 \n",
      " batch_normalization_3 (Batc  (None, 53, 53, 15)       60        \n",
      " hNormalization)                                                 \n",
      "                                                                 \n",
      " max_pooling2d_2 (MaxPooling  (None, 26, 26, 15)       0         \n",
      " 2D)                                                             \n",
      "                                                                 \n",
      " flatten (Flatten)           (None, 10140)             0         \n",
      "                                                                 \n",
      " dense (Dense)               (None, 100)               1014100   \n",
      "                                                                 \n",
      " dropout_2 (Dropout)         (None, 100)               0         \n",
      "                                                                 \n",
      " dense_1 (Dense)             (None, 2)                 202       \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 1,021,082\n",
      "Trainable params: 1,020,962\n",
      "Non-trainable params: 120\n",
      "_________________________________________________________________\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Kannan-PC\\anaconda3\\lib\\site-packages\\keras\\optimizers\\optimizer_v2\\adam.py:110: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.\n",
      "  super(Adam, self).__init__(name, **kwargs)\n"
     ]
    }
   ],
   "source": [
    "model1 = None\n",
    "model1 = createModel()\n",
    "batch_size = 40\n",
    "epochs = 150\n",
    "opt = optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.99, epsilon=None, decay=0)\n",
    "model1.compile(loss='binary_crossentropy', optimizer=opt,metrics = ['categorical_accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The model is fit to the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/150\n",
      "58/58 [==============================] - 142s 2s/step - loss: 0.6681 - categorical_accuracy: 0.7533 - val_loss: 0.8232 - val_categorical_accuracy: 0.5501\n",
      "Epoch 2/150\n",
      "58/58 [==============================] - 151s 3s/step - loss: 0.4124 - categorical_accuracy: 0.7992 - val_loss: 1.4596 - val_categorical_accuracy: 0.5905\n",
      "Epoch 3/150\n",
      "58/58 [==============================] - 139s 2s/step - loss: 0.3905 - categorical_accuracy: 0.8088 - val_loss: 1.1373 - val_categorical_accuracy: 0.6562\n",
      "Epoch 4/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.3530 - categorical_accuracy: 0.8105 - val_loss: 0.6749 - val_categorical_accuracy: 0.7634\n",
      "Epoch 5/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.3530 - categorical_accuracy: 0.8222 - val_loss: 0.8719 - val_categorical_accuracy: 0.7492\n",
      "Epoch 6/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.3322 - categorical_accuracy: 0.8248 - val_loss: 0.7086 - val_categorical_accuracy: 0.7897\n",
      "Epoch 7/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.3050 - categorical_accuracy: 0.8391 - val_loss: 0.5555 - val_categorical_accuracy: 0.7907\n",
      "Epoch 8/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.3087 - categorical_accuracy: 0.8365 - val_loss: 1.1197 - val_categorical_accuracy: 0.7169\n",
      "Epoch 9/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.2884 - categorical_accuracy: 0.8408 - val_loss: 0.5494 - val_categorical_accuracy: 0.7917\n",
      "Epoch 10/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.2741 - categorical_accuracy: 0.8565 - val_loss: 0.8898 - val_categorical_accuracy: 0.7644\n",
      "Epoch 11/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.2661 - categorical_accuracy: 0.8669 - val_loss: 0.5104 - val_categorical_accuracy: 0.7947\n",
      "Epoch 12/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.2482 - categorical_accuracy: 0.8695 - val_loss: 0.7626 - val_categorical_accuracy: 0.7351\n",
      "Epoch 13/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.2396 - categorical_accuracy: 0.8825 - val_loss: 1.2781 - val_categorical_accuracy: 0.6744\n",
      "Epoch 14/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.2186 - categorical_accuracy: 0.8903 - val_loss: 0.4971 - val_categorical_accuracy: 0.8119\n",
      "Epoch 15/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.2185 - categorical_accuracy: 0.8946 - val_loss: 0.5665 - val_categorical_accuracy: 0.7765\n",
      "Epoch 16/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.1854 - categorical_accuracy: 0.9068 - val_loss: 0.6975 - val_categorical_accuracy: 0.7583\n",
      "Epoch 17/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.2007 - categorical_accuracy: 0.9068 - val_loss: 0.7386 - val_categorical_accuracy: 0.7735\n",
      "Epoch 18/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1648 - categorical_accuracy: 0.9241 - val_loss: 0.5529 - val_categorical_accuracy: 0.7887\n",
      "Epoch 19/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1552 - categorical_accuracy: 0.9232 - val_loss: 1.0008 - val_categorical_accuracy: 0.7007\n",
      "Epoch 20/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1363 - categorical_accuracy: 0.9350 - val_loss: 0.7262 - val_categorical_accuracy: 0.7604\n",
      "Epoch 21/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1272 - categorical_accuracy: 0.9454 - val_loss: 0.9668 - val_categorical_accuracy: 0.7553\n",
      "Epoch 22/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1357 - categorical_accuracy: 0.9397 - val_loss: 1.0672 - val_categorical_accuracy: 0.7513\n",
      "Epoch 23/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1249 - categorical_accuracy: 0.9475 - val_loss: 0.7647 - val_categorical_accuracy: 0.8190\n",
      "Epoch 24/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1018 - categorical_accuracy: 0.9562 - val_loss: 0.9450 - val_categorical_accuracy: 0.7594\n",
      "Epoch 25/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.1047 - categorical_accuracy: 0.9584 - val_loss: 0.8343 - val_categorical_accuracy: 0.7998\n",
      "Epoch 26/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0938 - categorical_accuracy: 0.9584 - val_loss: 3.1970 - val_categorical_accuracy: 0.7422\n",
      "Epoch 27/150\n",
      "58/58 [==============================] - 1641s 29s/step - loss: 0.0970 - categorical_accuracy: 0.9640 - val_loss: 1.2984 - val_categorical_accuracy: 0.7988\n",
      "Epoch 28/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0758 - categorical_accuracy: 0.9670 - val_loss: 1.0844 - val_categorical_accuracy: 0.7806\n",
      "Epoch 29/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0723 - categorical_accuracy: 0.9683 - val_loss: 0.9550 - val_categorical_accuracy: 0.8190\n",
      "Epoch 30/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0684 - categorical_accuracy: 0.9718 - val_loss: 1.1128 - val_categorical_accuracy: 0.7978\n",
      "Epoch 31/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0635 - categorical_accuracy: 0.9757 - val_loss: 0.9511 - val_categorical_accuracy: 0.8089\n",
      "Epoch 32/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0542 - categorical_accuracy: 0.9775 - val_loss: 2.0134 - val_categorical_accuracy: 0.7867\n",
      "Epoch 33/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0568 - categorical_accuracy: 0.9775 - val_loss: 1.4063 - val_categorical_accuracy: 0.7765\n",
      "Epoch 34/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0558 - categorical_accuracy: 0.9783 - val_loss: 1.2930 - val_categorical_accuracy: 0.7846\n",
      "Epoch 35/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0427 - categorical_accuracy: 0.9831 - val_loss: 1.0449 - val_categorical_accuracy: 0.8069\n",
      "Epoch 36/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0669 - categorical_accuracy: 0.9735 - val_loss: 1.0028 - val_categorical_accuracy: 0.8079\n",
      "Epoch 37/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0657 - categorical_accuracy: 0.9744 - val_loss: 1.6238 - val_categorical_accuracy: 0.7735\n",
      "Epoch 38/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0394 - categorical_accuracy: 0.9840 - val_loss: 1.7288 - val_categorical_accuracy: 0.7735\n",
      "Epoch 39/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0525 - categorical_accuracy: 0.9801 - val_loss: 1.3820 - val_categorical_accuracy: 0.7917\n",
      "Epoch 40/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0415 - categorical_accuracy: 0.9853 - val_loss: 1.8030 - val_categorical_accuracy: 0.7978\n",
      "Epoch 41/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0494 - categorical_accuracy: 0.9818 - val_loss: 1.1158 - val_categorical_accuracy: 0.8069\n",
      "Epoch 42/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0626 - categorical_accuracy: 0.9792 - val_loss: 2.5294 - val_categorical_accuracy: 0.7988\n",
      "Epoch 43/150\n",
      "58/58 [==============================] - 43438s 762s/step - loss: 0.0480 - categorical_accuracy: 0.9809 - val_loss: 1.1121 - val_categorical_accuracy: 0.7897\n",
      "Epoch 44/150\n",
      "58/58 [==============================] - 133s 2s/step - loss: 0.0413 - categorical_accuracy: 0.9844 - val_loss: 1.1489 - val_categorical_accuracy: 0.8059\n",
      "Epoch 45/150\n",
      "58/58 [==============================] - 126s 2s/step - loss: 0.0485 - categorical_accuracy: 0.9831 - val_loss: 2.6603 - val_categorical_accuracy: 0.7240\n",
      "Epoch 46/150\n",
      "58/58 [==============================] - 126s 2s/step - loss: 0.0445 - categorical_accuracy: 0.9840 - val_loss: 1.0150 - val_categorical_accuracy: 0.8099\n",
      "Epoch 47/150\n",
      "58/58 [==============================] - 126s 2s/step - loss: 0.0416 - categorical_accuracy: 0.9840 - val_loss: 1.5193 - val_categorical_accuracy: 0.7796\n",
      "Epoch 48/150\n",
      "58/58 [==============================] - 126s 2s/step - loss: 0.0303 - categorical_accuracy: 0.9892 - val_loss: 1.3408 - val_categorical_accuracy: 0.7887\n",
      "Epoch 49/150\n",
      "58/58 [==============================] - 125s 2s/step - loss: 0.0396 - categorical_accuracy: 0.9866 - val_loss: 1.4906 - val_categorical_accuracy: 0.8150\n",
      "Epoch 50/150\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "58/58 [==============================] - 127s 2s/step - loss: 0.0398 - categorical_accuracy: 0.9857 - val_loss: 1.3159 - val_categorical_accuracy: 0.7786\n",
      "Epoch 51/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0582 - categorical_accuracy: 0.9831 - val_loss: 1.1189 - val_categorical_accuracy: 0.7776\n",
      "Epoch 52/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0577 - categorical_accuracy: 0.9818 - val_loss: 3.3963 - val_categorical_accuracy: 0.6866\n",
      "Epoch 53/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0446 - categorical_accuracy: 0.9861 - val_loss: 0.9831 - val_categorical_accuracy: 0.8311\n",
      "Epoch 54/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0238 - categorical_accuracy: 0.9883 - val_loss: 1.4979 - val_categorical_accuracy: 0.7958\n",
      "Epoch 55/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0204 - categorical_accuracy: 0.9887 - val_loss: 0.9677 - val_categorical_accuracy: 0.8402\n",
      "Epoch 56/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0302 - categorical_accuracy: 0.9892 - val_loss: 2.3160 - val_categorical_accuracy: 0.7300\n",
      "Epoch 57/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0451 - categorical_accuracy: 0.9857 - val_loss: 1.5754 - val_categorical_accuracy: 0.7846\n",
      "Epoch 58/150\n",
      "58/58 [==============================] - 11337s 199s/step - loss: 0.0430 - categorical_accuracy: 0.9892 - val_loss: 1.8809 - val_categorical_accuracy: 0.7260\n",
      "Epoch 59/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0300 - categorical_accuracy: 0.9892 - val_loss: 1.9538 - val_categorical_accuracy: 0.7128\n",
      "Epoch 60/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0288 - categorical_accuracy: 0.9896 - val_loss: 2.2820 - val_categorical_accuracy: 0.7017\n",
      "Epoch 61/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0423 - categorical_accuracy: 0.9870 - val_loss: 1.7817 - val_categorical_accuracy: 0.7614\n",
      "Epoch 62/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0479 - categorical_accuracy: 0.9835 - val_loss: 1.9238 - val_categorical_accuracy: 0.7037\n",
      "Epoch 63/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0241 - categorical_accuracy: 0.9909 - val_loss: 1.6399 - val_categorical_accuracy: 0.7543\n",
      "Epoch 64/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0246 - categorical_accuracy: 0.9896 - val_loss: 1.1465 - val_categorical_accuracy: 0.7937\n",
      "Epoch 65/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0280 - categorical_accuracy: 0.9892 - val_loss: 1.1783 - val_categorical_accuracy: 0.8140\n",
      "Epoch 66/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0270 - categorical_accuracy: 0.9918 - val_loss: 1.2319 - val_categorical_accuracy: 0.8028\n",
      "Epoch 67/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0131 - categorical_accuracy: 0.9948 - val_loss: 1.6859 - val_categorical_accuracy: 0.7412\n",
      "Epoch 68/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0182 - categorical_accuracy: 0.9948 - val_loss: 1.4405 - val_categorical_accuracy: 0.8028\n",
      "Epoch 69/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0366 - categorical_accuracy: 0.9883 - val_loss: 2.9715 - val_categorical_accuracy: 0.6936\n",
      "Epoch 70/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0502 - categorical_accuracy: 0.9870 - val_loss: 1.4246 - val_categorical_accuracy: 0.7968\n",
      "Epoch 71/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0266 - categorical_accuracy: 0.9909 - val_loss: 1.8726 - val_categorical_accuracy: 0.7776\n",
      "Epoch 72/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0202 - categorical_accuracy: 0.9935 - val_loss: 1.7341 - val_categorical_accuracy: 0.7644\n",
      "Epoch 73/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0181 - categorical_accuracy: 0.9935 - val_loss: 2.1613 - val_categorical_accuracy: 0.7776\n",
      "Epoch 74/150\n",
      "58/58 [==============================] - 11504s 202s/step - loss: 0.0216 - categorical_accuracy: 0.9918 - val_loss: 1.3442 - val_categorical_accuracy: 0.7927\n",
      "Epoch 75/150\n",
      "58/58 [==============================] - 129s 2s/step - loss: 0.0163 - categorical_accuracy: 0.9926 - val_loss: 1.3072 - val_categorical_accuracy: 0.7867\n",
      "Epoch 76/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0167 - categorical_accuracy: 0.9948 - val_loss: 1.5350 - val_categorical_accuracy: 0.7685\n",
      "Epoch 77/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0254 - categorical_accuracy: 0.9905 - val_loss: 1.5811 - val_categorical_accuracy: 0.7897\n",
      "Epoch 78/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0283 - categorical_accuracy: 0.9900 - val_loss: 1.4335 - val_categorical_accuracy: 0.7533\n",
      "Epoch 79/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0368 - categorical_accuracy: 0.9913 - val_loss: 1.6878 - val_categorical_accuracy: 0.7695\n",
      "Epoch 80/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0305 - categorical_accuracy: 0.9931 - val_loss: 1.1378 - val_categorical_accuracy: 0.8210\n",
      "Epoch 81/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0133 - categorical_accuracy: 0.9935 - val_loss: 1.7106 - val_categorical_accuracy: 0.7897\n",
      "Epoch 82/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0181 - categorical_accuracy: 0.9935 - val_loss: 2.1534 - val_categorical_accuracy: 0.7614\n",
      "Epoch 83/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0177 - categorical_accuracy: 0.9931 - val_loss: 2.0772 - val_categorical_accuracy: 0.7492\n",
      "Epoch 84/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0484 - categorical_accuracy: 0.9853 - val_loss: 2.3999 - val_categorical_accuracy: 0.7513\n",
      "Epoch 85/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0232 - categorical_accuracy: 0.9931 - val_loss: 0.8724 - val_categorical_accuracy: 0.8311\n",
      "Epoch 86/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0134 - categorical_accuracy: 0.9957 - val_loss: 1.2041 - val_categorical_accuracy: 0.8069\n",
      "Epoch 87/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0269 - categorical_accuracy: 0.9922 - val_loss: 1.1856 - val_categorical_accuracy: 0.8170\n",
      "Epoch 88/150\n",
      "58/58 [==============================] - 8218s 144s/step - loss: 0.0225 - categorical_accuracy: 0.9926 - val_loss: 1.9963 - val_categorical_accuracy: 0.7937\n",
      "Epoch 89/150\n",
      "58/58 [==============================] - 1464s 26s/step - loss: 0.0112 - categorical_accuracy: 0.9948 - val_loss: 1.1197 - val_categorical_accuracy: 0.8200\n",
      "Epoch 90/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0370 - categorical_accuracy: 0.9900 - val_loss: 1.4701 - val_categorical_accuracy: 0.7219\n",
      "Epoch 91/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0250 - categorical_accuracy: 0.9944 - val_loss: 1.6733 - val_categorical_accuracy: 0.7796\n",
      "Epoch 92/150\n",
      "58/58 [==============================] - 130s 2s/step - loss: 0.0212 - categorical_accuracy: 0.9935 - val_loss: 2.5267 - val_categorical_accuracy: 0.7351\n",
      "Epoch 93/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0316 - categorical_accuracy: 0.9918 - val_loss: 1.8550 - val_categorical_accuracy: 0.7685\n",
      "Epoch 94/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0154 - categorical_accuracy: 0.9948 - val_loss: 1.8560 - val_categorical_accuracy: 0.7786\n",
      "Epoch 95/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0144 - categorical_accuracy: 0.9939 - val_loss: 3.0552 - val_categorical_accuracy: 0.6896\n",
      "Epoch 96/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0138 - categorical_accuracy: 0.9965 - val_loss: 1.6546 - val_categorical_accuracy: 0.7978\n",
      "Epoch 97/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0253 - categorical_accuracy: 0.9931 - val_loss: 1.8368 - val_categorical_accuracy: 0.7523\n",
      "Epoch 98/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0135 - categorical_accuracy: 0.9961 - val_loss: 1.2827 - val_categorical_accuracy: 0.7816\n",
      "Epoch 99/150\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "58/58 [==============================] - 128s 2s/step - loss: 0.0217 - categorical_accuracy: 0.9922 - val_loss: 1.3614 - val_categorical_accuracy: 0.8109\n",
      "Epoch 100/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0275 - categorical_accuracy: 0.9918 - val_loss: 1.4200 - val_categorical_accuracy: 0.7735\n",
      "Epoch 101/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0188 - categorical_accuracy: 0.9939 - val_loss: 1.1317 - val_categorical_accuracy: 0.8059\n",
      "Epoch 102/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0077 - categorical_accuracy: 0.9974 - val_loss: 1.3692 - val_categorical_accuracy: 0.8018\n",
      "Epoch 103/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0168 - categorical_accuracy: 0.9948 - val_loss: 1.5036 - val_categorical_accuracy: 0.7391\n",
      "Epoch 104/150\n",
      "58/58 [==============================] - 16567s 291s/step - loss: 0.0175 - categorical_accuracy: 0.9948 - val_loss: 1.1627 - val_categorical_accuracy: 0.8160\n",
      "Epoch 105/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0100 - categorical_accuracy: 0.9957 - val_loss: 1.3068 - val_categorical_accuracy: 0.7998\n",
      "Epoch 106/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0192 - categorical_accuracy: 0.9939 - val_loss: 2.5328 - val_categorical_accuracy: 0.7776\n",
      "Epoch 107/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0322 - categorical_accuracy: 0.9900 - val_loss: 1.0785 - val_categorical_accuracy: 0.8008\n",
      "Epoch 108/150\n",
      "58/58 [==============================] - 130s 2s/step - loss: 0.0131 - categorical_accuracy: 0.9944 - val_loss: 1.4237 - val_categorical_accuracy: 0.7705\n",
      "Epoch 109/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0201 - categorical_accuracy: 0.9961 - val_loss: 1.2245 - val_categorical_accuracy: 0.8109\n",
      "Epoch 110/150\n",
      "58/58 [==============================] - 146s 3s/step - loss: 0.0276 - categorical_accuracy: 0.9896 - val_loss: 1.2333 - val_categorical_accuracy: 0.7877\n",
      "Epoch 111/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0181 - categorical_accuracy: 0.9935 - val_loss: 1.2738 - val_categorical_accuracy: 0.8322\n",
      "Epoch 112/150\n",
      "58/58 [==============================] - 145s 3s/step - loss: 0.0181 - categorical_accuracy: 0.9957 - val_loss: 1.1844 - val_categorical_accuracy: 0.8311\n",
      "Epoch 113/150\n",
      "58/58 [==============================] - 138s 2s/step - loss: 0.0123 - categorical_accuracy: 0.9952 - val_loss: 1.1314 - val_categorical_accuracy: 0.7947\n",
      "Epoch 114/150\n",
      "58/58 [==============================] - 144s 2s/step - loss: 0.0139 - categorical_accuracy: 0.9935 - val_loss: 1.0817 - val_categorical_accuracy: 0.7978\n",
      "Epoch 115/150\n",
      "58/58 [==============================] - 135s 2s/step - loss: 0.0181 - categorical_accuracy: 0.9926 - val_loss: 2.0763 - val_categorical_accuracy: 0.7907\n",
      "Epoch 116/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0189 - categorical_accuracy: 0.9957 - val_loss: 1.0228 - val_categorical_accuracy: 0.8089\n",
      "Epoch 117/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0171 - categorical_accuracy: 0.9931 - val_loss: 1.7531 - val_categorical_accuracy: 0.7725\n",
      "Epoch 118/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0210 - categorical_accuracy: 0.9935 - val_loss: 1.2252 - val_categorical_accuracy: 0.7604\n",
      "Epoch 119/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0136 - categorical_accuracy: 0.9957 - val_loss: 3.3229 - val_categorical_accuracy: 0.7088\n",
      "Epoch 120/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0095 - categorical_accuracy: 0.9970 - val_loss: 1.3754 - val_categorical_accuracy: 0.7937\n",
      "Epoch 121/150\n",
      "58/58 [==============================] - 137s 2s/step - loss: 0.0103 - categorical_accuracy: 0.9974 - val_loss: 1.9968 - val_categorical_accuracy: 0.8038\n",
      "Epoch 122/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0179 - categorical_accuracy: 0.9944 - val_loss: 1.7701 - val_categorical_accuracy: 0.7695\n",
      "Epoch 123/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0294 - categorical_accuracy: 0.9913 - val_loss: 1.4656 - val_categorical_accuracy: 0.7432\n",
      "Epoch 124/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0121 - categorical_accuracy: 0.9952 - val_loss: 1.5892 - val_categorical_accuracy: 0.8069\n",
      "Epoch 125/150\n",
      "58/58 [==============================] - 140s 2s/step - loss: 0.0105 - categorical_accuracy: 0.9974 - val_loss: 1.6552 - val_categorical_accuracy: 0.7856\n",
      "Epoch 126/150\n",
      "58/58 [==============================] - 136s 2s/step - loss: 0.0072 - categorical_accuracy: 0.9978 - val_loss: 1.5939 - val_categorical_accuracy: 0.8049\n",
      "Epoch 127/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0127 - categorical_accuracy: 0.9944 - val_loss: 1.4023 - val_categorical_accuracy: 0.8200\n",
      "Epoch 128/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0145 - categorical_accuracy: 0.9948 - val_loss: 1.8099 - val_categorical_accuracy: 0.7472\n",
      "Epoch 129/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0066 - categorical_accuracy: 0.9978 - val_loss: 1.3950 - val_categorical_accuracy: 0.8352\n",
      "Epoch 130/150\n",
      "58/58 [==============================] - 141s 2s/step - loss: 0.0074 - categorical_accuracy: 0.9965 - val_loss: 1.0897 - val_categorical_accuracy: 0.8291\n",
      "Epoch 131/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0052 - categorical_accuracy: 0.9974 - val_loss: 1.9949 - val_categorical_accuracy: 0.7877\n",
      "Epoch 132/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0150 - categorical_accuracy: 0.9965 - val_loss: 0.9798 - val_categorical_accuracy: 0.8008\n",
      "Epoch 133/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0277 - categorical_accuracy: 0.9926 - val_loss: 1.1459 - val_categorical_accuracy: 0.7776\n",
      "Epoch 134/150\n",
      "58/58 [==============================] - 132s 2s/step - loss: 0.0461 - categorical_accuracy: 0.9918 - val_loss: 1.5991 - val_categorical_accuracy: 0.8160\n",
      "Epoch 135/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0158 - categorical_accuracy: 0.9957 - val_loss: 1.5366 - val_categorical_accuracy: 0.8150\n",
      "Epoch 136/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0136 - categorical_accuracy: 0.9970 - val_loss: 0.9283 - val_categorical_accuracy: 0.7553\n",
      "Epoch 137/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0060 - categorical_accuracy: 0.9978 - val_loss: 1.5269 - val_categorical_accuracy: 0.8281\n",
      "Epoch 138/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0112 - categorical_accuracy: 0.9957 - val_loss: 0.8504 - val_categorical_accuracy: 0.8038\n",
      "Epoch 139/150\n",
      "58/58 [==============================] - 139s 2s/step - loss: 0.0209 - categorical_accuracy: 0.9935 - val_loss: 1.6417 - val_categorical_accuracy: 0.7735\n",
      "Epoch 140/150\n",
      "58/58 [==============================] - 138s 2s/step - loss: 0.0100 - categorical_accuracy: 0.9970 - val_loss: 1.2985 - val_categorical_accuracy: 0.8301\n",
      "Epoch 141/150\n",
      "58/58 [==============================] - 131s 2s/step - loss: 0.0120 - categorical_accuracy: 0.9961 - val_loss: 1.2448 - val_categorical_accuracy: 0.8190\n",
      "Epoch 142/150\n",
      "58/58 [==============================] - 134s 2s/step - loss: 0.0274 - categorical_accuracy: 0.9931 - val_loss: 1.4419 - val_categorical_accuracy: 0.8099\n",
      "Epoch 143/150\n",
      "58/58 [==============================] - 134s 2s/step - loss: 0.0321 - categorical_accuracy: 0.9931 - val_loss: 1.2034 - val_categorical_accuracy: 0.6775\n",
      "Epoch 144/150\n",
      "58/58 [==============================] - 129s 2s/step - loss: 0.0124 - categorical_accuracy: 0.9961 - val_loss: 1.0797 - val_categorical_accuracy: 0.8109\n",
      "Epoch 145/150\n",
      "58/58 [==============================] - 129s 2s/step - loss: 0.0083 - categorical_accuracy: 0.9965 - val_loss: 2.4201 - val_categorical_accuracy: 0.7169\n",
      "Epoch 146/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0103 - categorical_accuracy: 0.9965 - val_loss: 1.4563 - val_categorical_accuracy: 0.8119\n",
      "Epoch 147/150\n",
      "58/58 [==============================] - 128s 2s/step - loss: 0.0071 - categorical_accuracy: 0.9983 - val_loss: 1.2337 - val_categorical_accuracy: 0.8241\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 148/150\n",
      "58/58 [==============================] - 127s 2s/step - loss: 0.0042 - categorical_accuracy: 0.9987 - val_loss: 1.3117 - val_categorical_accuracy: 0.8311\n",
      "Epoch 149/150\n",
      "58/58 [==============================] - 139s 2s/step - loss: 0.0074 - categorical_accuracy: 0.9978 - val_loss: 1.9530 - val_categorical_accuracy: 0.8119\n",
      "Epoch 150/150\n",
      "58/58 [==============================] - 130s 2s/step - loss: 0.0109 - categorical_accuracy: 0.9961 - val_loss: 1.3061 - val_categorical_accuracy: 0.7391\n"
     ]
    }
   ],
   "source": [
    "history=model1.fit(x_train, y_train, epochs=150,batch_size = batch_size,validation_data= (x_Test, y_Test), shuffle=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The accuracy and loss graphs are obtained"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['categorical_accuracy'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train'], loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The classification report of the test data is printed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "31/31 [==============================] - 8s 253ms/step\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.79      0.71      0.75       540\n",
      "           1       0.69      0.78      0.73       449\n",
      "\n",
      "    accuracy                           0.74       989\n",
      "   macro avg       0.74      0.74      0.74       989\n",
      "weighted avg       0.75      0.74      0.74       989\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "pred = model.predict(x_Test, batch_size=32, verbose=1)\n",
    "predicted = np.argmax(pred, axis=1)\n",
    "report = classification_report(np.argmax(y_Test, axis=1), predicted)\n",
    "print(report)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After observing the CPVS of the test samples and the respective outputs, the threshold is calculated as 99.99995 and the BSD is created"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "import numpy as np\n",
    "import os \n",
    "for i in range(989):\n",
    "    if abs((pred[i][0]*100) - (pred[i][1]*100)) < 99.99995 :\n",
    "        img = Image.fromarray(x_Test[i], 'RGB')\n",
    "        if np.argmax(y_Test[i])== 0:\n",
    "                img.save(os.path.join('final/benign/',str(i)+'.jpg'))\n",
    "        if np.argmax(y_Test[i])== 1:\n",
    "                img.save(os.path.join('final/malignant/', str(i)+'.jpg'))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}