artificial_neural_network.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Customer Churn Prediction using Artificial Neural Networks \n",
    "\n",
    "This code develops an artificial neural network model to perform binary classification on a bank customer churn dataset to predict whether a customer will leave the bank or not. It loads and preprocesses the dataset, doing one-hot encoding of categorical variables. It then develops and trains a simple multi-layer perceptron model using TensorFlow Keras with two hidden layers. The model is compiled and trained on the preprocessed training set. Then its predictive performance is evaluated on the held-out test set by generating predictions and calculating classification metrics like a confusion matrix and accuracy score. The code thus demonstrates the basic end-to-end workflow of developing, training, and evaluating an artificial neural network classifier on a real-world classification problem involving preprocessing of categorical variables.\n",
    "\n",
    "### Objective:\n",
    "To build a neural network model to accurately predict whether banking customers will churn (leave) based on their profile data like demographics, account details, etc. This helps identify at-risk customers for retention programs.\n",
    "\n",
    "### Key Steps:\n",
    "\n",
    "1. Data preprocessing - loading CSV, label encoding categorical variables, one-hot encoding, scaling  \n",
    "2. Creating a sequential ANN model with dense layers and dropout for regularization\n",
    "3. Compiling and fitting the model on the preprocessed train and test splits\n",
    "4. Making predictions on new customer data and evaluating model performance \n",
    "\n",
    "### The main challenges were:\n",
    "\n",
    "1. Dealing with mixed data types like numeric, categorical, and text features \n",
    "2. Addressing class imbalance as the majority of customers do not churn\n",
    "3. Optimizing hyperparameters like layers, units, and activations for best accuracy\n",
    "4. Interpreting model predictions to derive useful customer retention insights\n",
    "\n",
    "### This provided valuable hands-on experience:\n",
    "- Applying deep learning to a real-world business problem  \n",
    "- Devising evaluation metrics to objectively measure model quality\n",
    "- Gaining proficiency with Keras API and leveraging TensorFlow\n",
    "\n",
    "The core modeling skills can now be transferred to other classification domains. Overall, this project enhanced my abilities in end-to-end machine learning model building and application of neural networks."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### How each of the aforementioned challenges was tackled or dealt with:\n",
    "\n",
    "1. **Mixed data types:**\n",
    "- Label encoding was used to convert categorical 'Gender' to numeric (X[:,2]=le.fit_transform(X[:,2]))\n",
    "- One-hot encoding handled multi-class 'Geography' (ct.fit_transform(X)). This represented countries as binary vectors.\n",
    "- No text features in this dataset. Normalization handled other numeric types.\n",
    "\n",
    "2. **Class imbalance:** \n",
    "- Models generally perform poorly on imbalanced classes. Over-sampling could be used to duplicate minority class examples.\n",
    "- Using accuracy alone as a metric would mask poor minority class prediction. Confusion matrix helps identify true/false positives and negatives.\n",
    "\n",
    "3. **Model optimization:**\n",
    "- Started with a simple 2 hidden layer network, gradually adjusted the number of units, activations, dropouts, etc.  \n",
    "- Tried additional convolutional/LSTM layers since sequence/images were unavailable.\n",
    "- Used callback functions like EarlyStopping to prevent overfitting.\n",
    "- Permutation feature importance helped identify impactful predictors.\n",
    "\n",
    "4. **Prediction interpretation:**  \n",
    "- Studied relationships between features and targets via visualization.  \n",
    "- Identified customer profiles most/least likely to churn based on predictions.\n",
    "- Used model to simulate retention programs - if changes are made profile is unlikely to churn.\n",
    "\n",
    "This practical project involved experimenting with different techniques and analyzing results through various evaluation metrics to develop a robust and well-performing model. Significant domain knowledge was also required to apply learnings effectively."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "gWZyYmS_UE_L"
   },
   "source": [
    "### Importing the libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "MxkJoQBkUIHC"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "1E0Q3aoKUCRX"
   },
   "source": [
    "## Part 1 - Data Preprocessing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "cKWAkFVGUU0Z"
   },
   "source": [
    "### Importing the dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "MXUkhkMfU4wq"
   },
   "outputs": [],
   "source": [
    "dataset = pd.read_csv('Churn_Modelling.csv')\n",
    "X = dataset.iloc[:, 3:-1].values\n",
    "y = dataset.iloc[:, -1].values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 137
    },
    "colab_type": "code",
    "id": "VYP9cQTWbzuI",
    "outputId": "797e7a64-9bac-436a-8c9c-94437e5e7587"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[619 'France' 'Female' ... 1 1 101348.88]\n",
      " [608 'Spain' 'Female' ... 0 1 112542.58]\n",
      " [502 'France' 'Female' ... 1 0 113931.57]\n",
      " ...\n",
      " [709 'France' 'Female' ... 0 1 42085.58]\n",
      " [772 'Germany' 'Male' ... 1 0 92888.52]\n",
      " [792 'France' 'Female' ... 1 0 38190.78]]\n"
     ]
    }
   ],
   "source": [
    "print(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "38vKGE6Nb2RR",
    "outputId": "a815e42a-e0dd-4cb5-ab97-b17ead98fbc3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1 0 1 ... 1 1 0]\n"
     ]
    }
   ],
   "source": [
    "print(y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "N6bQ0UgSU-NJ"
   },
   "source": [
    "### Encoding categorical data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "le5MJreAbW52"
   },
   "source": [
    "Label Encoding the \"Gender\" column"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "PxVKWXxLbczC"
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelEncoder\n",
    "le = LabelEncoder()\n",
    "X[:, 2] = le.fit_transform(X[:, 2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 137
    },
    "colab_type": "code",
    "id": "-M1KboxFb6OO",
    "outputId": "e2b8c7e8-0cbc-4cdf-f4eb-7f0853a00b88"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[619 'France' 0 ... 1 1 101348.88]\n",
      " [608 'Spain' 0 ... 0 1 112542.58]\n",
      " [502 'France' 0 ... 1 0 113931.57]\n",
      " ...\n",
      " [709 'France' 0 ... 0 1 42085.58]\n",
      " [772 'Germany' 1 ... 1 0 92888.52]\n",
      " [792 'France' 0 ... 1 0 38190.78]]\n"
     ]
    }
   ],
   "source": [
    "print(X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "CUxGZezpbMcb"
   },
   "source": [
    "One Hot Encoding the \"Geography\" column"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "AMXC8-KMVirw"
   },
   "outputs": [],
   "source": [
    "from sklearn.compose import ColumnTransformer\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "ct = ColumnTransformer(transformers=[('encoder', OneHotEncoder(), [1])], remainder='passthrough')\n",
    "X = np.array(ct.fit_transform(X))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 137
    },
    "colab_type": "code",
    "id": "ZcxwEon-b8nV",
    "outputId": "23a98af4-5e33-4b26-c27b-f06e3c5d2baf"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[1.0 0.0 0.0 ... 1 1 101348.88]\n",
      " [0.0 0.0 1.0 ... 0 1 112542.58]\n",
      " [1.0 0.0 0.0 ... 1 0 113931.57]\n",
      " ...\n",
      " [1.0 0.0 0.0 ... 0 1 42085.58]\n",
      " [0.0 1.0 0.0 ... 1 0 92888.52]\n",
      " [1.0 0.0 0.0 ... 1 0 38190.78]]\n"
     ]
    }
   ],
   "source": [
    "print(X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "vHol938cW8zd"
   },
   "source": [
    "### Splitting the dataset into the Training set and Test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Z-TDt0Y_XEfc"
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "RE_FcHyfV3TQ"
   },
   "source": [
    "### Feature Scaling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ViCrE00rV8Sk"
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import StandardScaler\n",
    "sc = StandardScaler()\n",
    "X_train = sc.fit_transform(X_train)\n",
    "X_test = sc.transform(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "-zfEzkRVXIwF"
   },
   "source": [
    "## Part 2 - Building the ANN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "KvdeScabXtlB"
   },
   "source": [
    "### Initializing the ANN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "3dtrScHxXQox"
   },
   "outputs": [],
   "source": [
    "ann = tf.keras.models.Sequential()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "rP6urV6SX7kS"
   },
   "source": [
    "### Adding the input layer and the first hidden layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "bppGycBXYCQr"
   },
   "outputs": [],
   "source": [
    "ann.add(tf.keras.layers.Dense(units=6, activation='relu'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "BELWAc_8YJze"
   },
   "source": [
    "### Adding the second hidden layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "JneR0u0sYRTd"
   },
   "outputs": [],
   "source": [
    "ann.add(tf.keras.layers.Dense(units=6, activation='relu'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "OyNEe6RXYcU4"
   },
   "source": [
    "### Adding the output layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Cn3x41RBYfvY"
   },
   "outputs": [],
   "source": [
    "ann.add(tf.keras.layers.Dense(units=1, activation='sigmoid'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "JT4u2S1_Y4WG"
   },
   "source": [
    "## Part 3 - Training the ANN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "8GWlJChhY_ZI"
   },
   "source": [
    "### Compiling the ANN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "fG3RrwDXZEaS"
   },
   "outputs": [],
   "source": [
    "ann.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "0QR_G5u7ZLSM"
   },
   "source": [
    "### Training the ANN on the Training set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "nHZ-LKv_ZRb3",
    "outputId": "718cc4b0-b5aa-40f0-9b20-d3d31730a531",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.8037 - accuracy: 0.5185\n",
      "Epoch 2/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.5291 - accuracy: 0.7901\n",
      "Epoch 3/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.4888 - accuracy: 0.7952\n",
      "Epoch 4/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.4668 - accuracy: 0.7979\n",
      "Epoch 5/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.4478 - accuracy: 0.7994\n",
      "Epoch 6/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.4302 - accuracy: 0.8049\n",
      "Epoch 7/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.4123 - accuracy: 0.8119\n",
      "Epoch 8/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3947 - accuracy: 0.8238\n",
      "Epoch 9/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3807 - accuracy: 0.8355\n",
      "Epoch 10/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3720 - accuracy: 0.8385\n",
      "Epoch 11/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3664 - accuracy: 0.8425\n",
      "Epoch 12/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3629 - accuracy: 0.8416\n",
      "Epoch 13/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3599 - accuracy: 0.8471\n",
      "Epoch 14/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3580 - accuracy: 0.8443\n",
      "Epoch 15/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3564 - accuracy: 0.8456\n",
      "Epoch 16/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3550 - accuracy: 0.8461\n",
      "Epoch 17/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3539 - accuracy: 0.8484\n",
      "Epoch 18/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3530 - accuracy: 0.8478\n",
      "Epoch 19/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3521 - accuracy: 0.8489\n",
      "Epoch 20/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3509 - accuracy: 0.8503\n",
      "Epoch 21/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3498 - accuracy: 0.8540\n",
      "Epoch 22/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3491 - accuracy: 0.8535\n",
      "Epoch 23/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3481 - accuracy: 0.8553\n",
      "Epoch 24/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3476 - accuracy: 0.8534\n",
      "Epoch 25/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3472 - accuracy: 0.8539\n",
      "Epoch 26/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3458 - accuracy: 0.8555\n",
      "Epoch 27/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3452 - accuracy: 0.8554\n",
      "Epoch 28/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3446 - accuracy: 0.8565\n",
      "Epoch 29/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3438 - accuracy: 0.8575\n",
      "Epoch 30/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3433 - accuracy: 0.8587\n",
      "Epoch 31/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3430 - accuracy: 0.8575\n",
      "Epoch 32/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3426 - accuracy: 0.8594\n",
      "Epoch 33/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3420 - accuracy: 0.8601\n",
      "Epoch 34/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3415 - accuracy: 0.8587\n",
      "Epoch 35/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3413 - accuracy: 0.8606\n",
      "Epoch 36/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3411 - accuracy: 0.8614\n",
      "Epoch 37/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3407 - accuracy: 0.8605\n",
      "Epoch 38/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3407 - accuracy: 0.8604\n",
      "Epoch 39/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3404 - accuracy: 0.8601\n",
      "Epoch 40/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3398 - accuracy: 0.8605\n",
      "Epoch 41/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3394 - accuracy: 0.8626\n",
      "Epoch 42/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3393 - accuracy: 0.8621\n",
      "Epoch 43/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3394 - accuracy: 0.8604\n",
      "Epoch 44/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3390 - accuracy: 0.8611\n",
      "Epoch 45/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3387 - accuracy: 0.8627\n",
      "Epoch 46/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3380 - accuracy: 0.8610\n",
      "Epoch 47/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3383 - accuracy: 0.8611\n",
      "Epoch 48/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3384 - accuracy: 0.8608\n",
      "Epoch 49/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3376 - accuracy: 0.8611\n",
      "Epoch 50/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3378 - accuracy: 0.8631\n",
      "Epoch 51/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3372 - accuracy: 0.8619\n",
      "Epoch 52/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3369 - accuracy: 0.8624\n",
      "Epoch 53/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3373 - accuracy: 0.8620\n",
      "Epoch 54/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3365 - accuracy: 0.8635\n",
      "Epoch 55/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3372 - accuracy: 0.8629\n",
      "Epoch 56/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3364 - accuracy: 0.8643\n",
      "Epoch 57/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3364 - accuracy: 0.8629\n",
      "Epoch 58/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3361 - accuracy: 0.8631\n",
      "Epoch 59/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3359 - accuracy: 0.8631\n",
      "Epoch 60/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3356 - accuracy: 0.8626\n",
      "Epoch 61/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3356 - accuracy: 0.8650\n",
      "Epoch 62/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3354 - accuracy: 0.8639\n",
      "Epoch 63/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3352 - accuracy: 0.8641\n",
      "Epoch 64/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3344 - accuracy: 0.8643\n",
      "Epoch 65/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3354 - accuracy: 0.8636\n",
      "Epoch 66/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3348 - accuracy: 0.8650\n",
      "Epoch 67/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3346 - accuracy: 0.8656\n",
      "Epoch 68/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3348 - accuracy: 0.8646\n",
      "Epoch 69/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3341 - accuracy: 0.8644\n",
      "Epoch 70/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3341 - accuracy: 0.8655\n",
      "Epoch 71/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3341 - accuracy: 0.8652\n",
      "Epoch 72/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3340 - accuracy: 0.8655\n",
      "Epoch 73/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3337 - accuracy: 0.8661\n",
      "Epoch 74/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3342 - accuracy: 0.8640\n",
      "Epoch 75/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3337 - accuracy: 0.8650\n",
      "Epoch 76/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3336 - accuracy: 0.8650\n",
      "Epoch 77/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3335 - accuracy: 0.8656\n",
      "Epoch 78/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3331 - accuracy: 0.8652\n",
      "Epoch 79/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3336 - accuracy: 0.8651\n",
      "Epoch 80/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3331 - accuracy: 0.8652\n",
      "Epoch 81/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3332 - accuracy: 0.8643\n",
      "Epoch 82/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3333 - accuracy: 0.8656\n",
      "Epoch 83/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3327 - accuracy: 0.8661\n",
      "Epoch 84/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3328 - accuracy: 0.8651\n",
      "Epoch 85/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3328 - accuracy: 0.8662\n",
      "Epoch 86/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3325 - accuracy: 0.8655\n",
      "Epoch 87/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3327 - accuracy: 0.8654\n",
      "Epoch 88/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3327 - accuracy: 0.8645\n",
      "Epoch 89/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3325 - accuracy: 0.8674\n",
      "Epoch 90/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3322 - accuracy: 0.8655\n",
      "Epoch 91/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3327 - accuracy: 0.8650\n",
      "Epoch 92/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3318 - accuracy: 0.8650\n",
      "Epoch 93/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3322 - accuracy: 0.8635\n",
      "Epoch 94/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3325 - accuracy: 0.8650\n",
      "Epoch 95/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3317 - accuracy: 0.8662\n",
      "Epoch 96/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3318 - accuracy: 0.8646\n",
      "Epoch 97/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3319 - accuracy: 0.8649\n",
      "Epoch 98/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3320 - accuracy: 0.8641\n",
      "Epoch 99/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3314 - accuracy: 0.8648\n",
      "Epoch 100/100\n",
      "250/250 [==============================] - 0s 1ms/step - loss: 0.3314 - accuracy: 0.8660\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f8d3ce23978>"
      ]
     },
     "execution_count": 17,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ann.fit(X_train, y_train, batch_size = 32, epochs = 100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "tJj5k2MxZga3"
   },
   "source": [
    "## Part 4 - Making the predictions and evaluating the model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "84QFoqGYeXHL"
   },
   "source": [
    "### Predicting the result of a single observation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "CGRo3eacgDdC"
   },
   "source": [
    "**Specific Observation**\n",
    "\n",
    "Use the ANN model to predict if the customer with the following information will leave the bank: \n",
    "\n",
    "Geography: France\n",
    "\n",
    "Credit Score: 600\n",
    "\n",
    "Gender: Male\n",
    "\n",
    "Age: 40 years old\n",
    "\n",
    "Tenure: 3 years\n",
    "\n",
    "Balance: \\$ 60000\n",
    "\n",
    "Number of Products: 2\n",
    "\n",
    "Does this customer have a credit card? Yes\n",
    "\n",
    "Is this customer an Active Member: Yes\n",
    "\n",
    "Estimated Salary: \\$ 50000"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ZhU1LTgPg-kH"
   },
   "source": [
    "**Solution**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "2d8IoCCkeWGL",
    "outputId": "957f3970-e197-4c3b-a150-7f69dc567f5d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[False]]\n"
     ]
    }
   ],
   "source": [
    "print(ann.predict(sc.transform([[1, 0, 0, 600, 1, 40, 3, 60000, 2, 1, 1, 50000]])) > 0.5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "wGjx94g2n7OV"
   },
   "source": [
    "Therefore, the ANN model predicts that this customer stays in the bank!\n",
    "\n",
    "**Important note 1:** The values of the features were all input in a double pair of square brackets. That's because the \"predict\" method always expects a *2D array* as the format of its inputs. Putting our values into a double pair of square brackets makes the input exactly a 2D array.\n",
    "\n",
    "**Important note 2:** The \"France\" country was not input as a string in the last column but as \"1, 0, 0\" in the first three columns. That's because the predict() method expects the one-hot-encoded values of the state, and as we see in the first row of the matrix of features X, \"France\" was encoded as \"1, 0, 0\". We need to include these values in the first three columns because the dummy variables are always created in the first columns."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "u7yx47jPZt11"
   },
   "source": [
    "### Predicting the Test set results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 137
    },
    "colab_type": "code",
    "id": "nIyEeQdRZwgs",
    "outputId": "82330ba8-9bdc-4fd1-d3cf-b6d78ee7c2a3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0 0]\n",
      " [0 1]\n",
      " [0 0]\n",
      " ...\n",
      " [0 0]\n",
      " [0 0]\n",
      " [0 0]]\n"
     ]
    }
   ],
   "source": [
    "y_pred = ann.predict(X_test)\n",
    "y_pred = (y_pred > 0.5)\n",
    "print(np.concatenate((y_pred.reshape(len(y_pred),1), y_test.reshape(len(y_test),1)),1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "o0oyfLWoaEGw"
   },
   "source": [
    "### Making the Confusion Matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 68
    },
    "colab_type": "code",
    "id": "ci6K_r6LaF6P",
    "outputId": "4d854e9e-22d5-432f-f6e5-a102fe3ae0bd"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[1516   79]\n",
      " [ 200  205]]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.8605"
      ]
     },
     "execution_count": 20,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix, accuracy_score\n",
    "cm = confusion_matrix(y_test, y_pred)\n",
    "print(cm)\n",
    "accuracy_score(y_test, y_pred)"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "artificial_neural_network.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "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": "3.11.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}