airplane_fatality.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "0400ddd5",
   "metadata": {},
   "source": [
    "# Importing Libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "a5a328bd",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "import requests\n",
    "from tqdm.notebook import tqdm\n",
    "import spacy\n",
    "nlp = spacy.load(\"en_core_web_sm\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e3cb670f",
   "metadata": {},
   "source": [
    "# Reading in the excel data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "8b63373a",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_excel('Airplane_Crashes_and_Fatalities_Since_1908.xlsx')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "1ca50f31",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Date</th>\n",
       "      <th>Time</th>\n",
       "      <th>Location</th>\n",
       "      <th>Operator</th>\n",
       "      <th>Flight #</th>\n",
       "      <th>Route</th>\n",
       "      <th>Type</th>\n",
       "      <th>Registration</th>\n",
       "      <th>cn/In</th>\n",
       "      <th>Aboard</th>\n",
       "      <th>Fatalities</th>\n",
       "      <th>Ground</th>\n",
       "      <th>Summary</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1908-09-17</td>\n",
       "      <td>17:18:00</td>\n",
       "      <td>Fort Myer, Virginia</td>\n",
       "      <td>Military - U.S. Army</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Demonstration</td>\n",
       "      <td>Wright Flyer III</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>During a demonstration flight, a U.S. Army fly...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1912-07-12</td>\n",
       "      <td>06:30:00</td>\n",
       "      <td>AtlantiCity, New Jersey</td>\n",
       "      <td>Military - U.S. Navy</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Test flight</td>\n",
       "      <td>Dirigible</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>5.0</td>\n",
       "      <td>5.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>First U.S. dirigible Akron exploded just offsh...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1913-08-06</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Victoria, British Columbia, Canada</td>\n",
       "      <td>Private</td>\n",
       "      <td>-</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Curtiss seaplane</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>The first fatal airplane accident in Canada oc...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1913-09-09</td>\n",
       "      <td>18:30:00</td>\n",
       "      <td>Over the North Sea</td>\n",
       "      <td>Military - German Navy</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Zeppelin L-1 (airship)</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>20.0</td>\n",
       "      <td>14.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>The airship flew into a thunderstorm and encou...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1913-10-17</td>\n",
       "      <td>10:30:00</td>\n",
       "      <td>Near Johannisthal, Germany</td>\n",
       "      <td>Military - German Navy</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Zeppelin L-2 (airship)</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>30.0</td>\n",
       "      <td>30.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Hydrogen gas which was being vented was sucked...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Date      Time                            Location  \\\n",
       "0 1908-09-17  17:18:00                 Fort Myer, Virginia   \n",
       "1 1912-07-12  06:30:00             AtlantiCity, New Jersey   \n",
       "2 1913-08-06       NaN  Victoria, British Columbia, Canada   \n",
       "3 1913-09-09  18:30:00                  Over the North Sea   \n",
       "4 1913-10-17  10:30:00          Near Johannisthal, Germany   \n",
       "\n",
       "                 Operator Flight #          Route                    Type  \\\n",
       "0    Military - U.S. Army      NaN  Demonstration        Wright Flyer III   \n",
       "1    Military - U.S. Navy      NaN    Test flight               Dirigible   \n",
       "2                 Private        -            NaN        Curtiss seaplane   \n",
       "3  Military - German Navy      NaN            NaN  Zeppelin L-1 (airship)   \n",
       "4  Military - German Navy      NaN            NaN  Zeppelin L-2 (airship)   \n",
       "\n",
       "  Registration cn/In  Aboard  Fatalities  Ground  \\\n",
       "0          NaN     1     2.0         1.0     0.0   \n",
       "1          NaN   NaN     5.0         5.0     0.0   \n",
       "2          NaN   NaN     1.0         1.0     0.0   \n",
       "3          NaN   NaN    20.0        14.0     0.0   \n",
       "4          NaN   NaN    30.0        30.0     0.0   \n",
       "\n",
       "                                             Summary  \n",
       "0  During a demonstration flight, a U.S. Army fly...  \n",
       "1  First U.S. dirigible Akron exploded just offsh...  \n",
       "2  The first fatal airplane accident in Canada oc...  \n",
       "3  The airship flew into a thunderstorm and encou...  \n",
       "4  Hydrogen gas which was being vented was sucked...  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "4286169f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 5268 entries, 0 to 5267\n",
      "Data columns (total 13 columns):\n",
      " #   Column        Non-Null Count  Dtype         \n",
      "---  ------        --------------  -----         \n",
      " 0   Date          5268 non-null   datetime64[ns]\n",
      " 1   Time          3049 non-null   object        \n",
      " 2   Location      5248 non-null   object        \n",
      " 3   Operator      5250 non-null   object        \n",
      " 4   Flight #      1069 non-null   object        \n",
      " 5   Route         3562 non-null   object        \n",
      " 6   Type          5241 non-null   object        \n",
      " 7   Registration  4933 non-null   object        \n",
      " 8   cn/In         4040 non-null   object        \n",
      " 9   Aboard        5246 non-null   float64       \n",
      " 10  Fatalities    5256 non-null   float64       \n",
      " 11  Ground        5246 non-null   float64       \n",
      " 12  Summary       4878 non-null   object        \n",
      "dtypes: datetime64[ns](1), float64(3), object(9)\n",
      "memory usage: 535.2+ KB\n"
     ]
    }
   ],
   "source": [
    "df.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "837851fd",
   "metadata": {},
   "source": [
    "# Creating a function to get the country code for the location of the crash"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "d1418568",
   "metadata": {},
   "outputs": [],
   "source": [
    "URL = \"https://geocoder.ls.hereapi.com/search/6.2/geocode.json\"\n",
    "api_key = 'XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX'  # Acquire api key from developer.here.com\n",
    "\n",
    "def getCoordinates(location):\n",
    "    PARAMS = {'searchtext': location, 'apikey': api_key} # required parameters\n",
    "    r = requests.get(url=URL, params=PARAMS)  # pass in required parameters\n",
    "    data = r.json() # Raw json file\n",
    "    try:\n",
    "        country = data['Response']['View'][0]['Result'][0]['Location']['Address']['Country']\n",
    "        return country\n",
    "    except:\n",
    "        return 'Unknown'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "656cc695",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = df.copy().reset_index()\n",
    "\n",
    "hereloc = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "7e56467e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "eb79deec63de44678a23068b73e33e61",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/2428 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#This would take at least 1 hr to run so you can use the incomplete.xlsx file and follow along\n",
    "for i, row in tqdm(data.iterrows(), total=len(data), miniters=1):\n",
    "    myid = row['index']\n",
    "    locate = row['Location']\n",
    "\n",
    "    country_code = getCoordinates(locate)\n",
    "    hereloc[myid] = [country_code]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "c467754e",
   "metadata": {},
   "outputs": [],
   "source": [
    "df['country_code'] = pd.DataFrame.from_dict(hereloc, orient='index')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0e8ae20d",
   "metadata": {},
   "source": [
    "Viewing the new dataframe"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 217,
   "id": "0ab35f98",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Date</th>\n",
       "      <th>Time</th>\n",
       "      <th>Location</th>\n",
       "      <th>Operator</th>\n",
       "      <th>Flight #</th>\n",
       "      <th>Route</th>\n",
       "      <th>Type</th>\n",
       "      <th>Registration</th>\n",
       "      <th>cn/In</th>\n",
       "      <th>Aboard</th>\n",
       "      <th>Fatalities</th>\n",
       "      <th>Ground</th>\n",
       "      <th>Summary</th>\n",
       "      <th>country_code</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1908-09-17</td>\n",
       "      <td>17:18:00</td>\n",
       "      <td>Fort Myer, Virginia</td>\n",
       "      <td>Military - U.S. Army</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Demonstration</td>\n",
       "      <td>Wright Flyer III</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1</td>\n",
       "      <td>2.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>During a demonstration flight, a U.S. Army fly...</td>\n",
       "      <td>USA</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1912-07-12</td>\n",
       "      <td>06:30:00</td>\n",
       "      <td>AtlantiCity, New Jersey</td>\n",
       "      <td>Military - U.S. Navy</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Test flight</td>\n",
       "      <td>Dirigible</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>5.0</td>\n",
       "      <td>5.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>First U.S. dirigible Akron exploded just offsh...</td>\n",
       "      <td>USA</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1913-08-06</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Victoria, British Columbia, Canada</td>\n",
       "      <td>Private</td>\n",
       "      <td>-</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Curtiss seaplane</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>The first fatal airplane accident in Canada oc...</td>\n",
       "      <td>CAN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1913-09-09</td>\n",
       "      <td>18:30:00</td>\n",
       "      <td>Over the North Sea</td>\n",
       "      <td>Military - German Navy</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Zeppelin L-1 (airship)</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>20.0</td>\n",
       "      <td>14.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>The airship flew into a thunderstorm and encou...</td>\n",
       "      <td>Unknown</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1913-10-17</td>\n",
       "      <td>10:30:00</td>\n",
       "      <td>Near Johannisthal, Germany</td>\n",
       "      <td>Military - German Navy</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Zeppelin L-2 (airship)</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>30.0</td>\n",
       "      <td>30.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>Hydrogen gas which was being vented was sucked...</td>\n",
       "      <td>DEU</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Date      Time                            Location  \\\n",
       "0 1908-09-17  17:18:00                 Fort Myer, Virginia   \n",
       "1 1912-07-12  06:30:00             AtlantiCity, New Jersey   \n",
       "2 1913-08-06       NaN  Victoria, British Columbia, Canada   \n",
       "3 1913-09-09  18:30:00                  Over the North Sea   \n",
       "4 1913-10-17  10:30:00          Near Johannisthal, Germany   \n",
       "\n",
       "                 Operator Flight #          Route                    Type  \\\n",
       "0    Military - U.S. Army      NaN  Demonstration        Wright Flyer III   \n",
       "1    Military - U.S. Navy      NaN    Test flight               Dirigible   \n",
       "2                 Private        -            NaN        Curtiss seaplane   \n",
       "3  Military - German Navy      NaN            NaN  Zeppelin L-1 (airship)   \n",
       "4  Military - German Navy      NaN            NaN  Zeppelin L-2 (airship)   \n",
       "\n",
       "  Registration cn/In  Aboard  Fatalities  Ground  \\\n",
       "0          NaN     1     2.0         1.0     0.0   \n",
       "1          NaN   NaN     5.0         5.0     0.0   \n",
       "2          NaN   NaN     1.0         1.0     0.0   \n",
       "3          NaN   NaN    20.0        14.0     0.0   \n",
       "4          NaN   NaN    30.0        30.0     0.0   \n",
       "\n",
       "                                             Summary country_code  \n",
       "0  During a demonstration flight, a U.S. Army fly...          USA  \n",
       "1  First U.S. dirigible Akron exploded just offsh...          USA  \n",
       "2  The first fatal airplane accident in Canada oc...          CAN  \n",
       "3  The airship flew into a thunderstorm and encou...      Unknown  \n",
       "4  Hydrogen gas which was being vented was sucked...          DEU  "
      ]
     },
     "execution_count": 217,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5363de9b",
   "metadata": {},
   "source": [
    "# Text Clustering"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e84ffec9",
   "metadata": {},
   "source": [
    "This function was to get the lower case root version of each word in the summary column"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 893,
   "id": "fe59ecb1",
   "metadata": {},
   "outputs": [],
   "source": [
    "def stem_punct_stop(word):\n",
    "    lt = []\n",
    "    for tokens in nlp(word):\n",
    "        if (tokens.is_punct) | (tokens.pos_ == 'NUM') | (tokens.is_stop):\n",
    "            #checking if the word is a punctuation, a number or a stopword\n",
    "            continue\n",
    "        lt.append(tokens.lemma_.lower())\n",
    "    return ' '.join(lt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 894,
   "id": "d345edde",
   "metadata": {},
   "outputs": [],
   "source": [
    "lem_text = df['Summary'].dropna().apply(stem_punct_stop)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d11083f",
   "metadata": {},
   "source": [
    "KMeans generally only works with numbers so we'll use an extraction tool to get numbers.\n",
    "\n",
    "The feature extraction tool used is TF-IDF, which is a numerical statistic. The term frequency and inverse document frequency are used in this statistic."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 895,
   "id": "e3660265",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.cluster import KMeans\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from kneed import KneeLocator #to locate the elbow (number of clusters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 896,
   "id": "dae1b96c",
   "metadata": {},
   "outputs": [],
   "source": [
    "text = list(lem_text)\n",
    "\n",
    "tv = TfidfVectorizer(stop_words='english')\n",
    "X = tv.fit_transform(text)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4f3694d7",
   "metadata": {},
   "source": [
    "Steps involved in choosing the number of clusters to use"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 902,
   "id": "1dc79578",
   "metadata": {},
   "outputs": [],
   "source": [
    "ses = []\n",
    "for k in range(1, 15):\n",
    "    kmeans = KMeans(n_clusters=k)\n",
    "    kmeans.fit(X)\n",
    "    ses.append(kmeans.inertia_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 903,
   "id": "7011b2bd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'SES')"
      ]
     },
     "execution_count": 903,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(range(1, 15), ses)\n",
    "plt.xticks(range(1, 15))\n",
    "plt.xlabel(\"Number of Clusters\")\n",
    "plt.ylabel(\"SES\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 904,
   "id": "8ff9d1d9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "6"
      ]
     },
     "execution_count": 904,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# To locate the elbow, i.e. the recommended number of clusters to use\n",
    "KneeLocator(range(1, 15), ses, curve=\"convex\", direction=\"decreasing\").elbow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 905,
   "id": "25320c18",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "KMeans(n_clusters=6, random_state=2)"
      ]
     },
     "execution_count": 905,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = KMeans(n_clusters=6, random_state=2)\n",
    "model.fit(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 906,
   "id": "6ba43568",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(6, 7669)"
      ]
     },
     "execution_count": 906,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.cluster_centers_.argsort().shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1174,
   "id": "7ea61f09",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Cluster 0\n",
      "en\n",
      "route\n",
      "crash\n",
      "mountain\n",
      "disappear\n",
      "plane\n",
      "cargo\n",
      "wreckage\n",
      "aircraft\n",
      "mountainous\n",
      "poor\n",
      "terrain\n",
      "weather\n",
      "strike\n",
      "miss\n",
      "mt\n",
      "undetermined\n",
      "flight\n",
      "sea\n",
      "mail\n",
      "\n",
      "\n",
      "Cluster 1\n",
      "approach\n",
      "runway\n",
      "crash\n",
      "land\n",
      "attempt\n",
      "plane\n",
      "short\n",
      "cargo\n",
      "aircraft\n",
      "landing\n",
      "fog\n",
      "strike\n",
      "tree\n",
      "pilot\n",
      "crew\n",
      "airport\n",
      "ground\n",
      "heavy\n",
      "altitude\n",
      "final\n",
      "\n",
      "\n",
      "Cluster 2\n",
      "crash\n",
      "aircraft\n",
      "plane\n",
      "pilot\n",
      "flight\n",
      "control\n",
      "fly\n",
      "sea\n",
      "crew\n",
      "cause\n",
      "ground\n",
      "shoot\n",
      "air\n",
      "altitude\n",
      "kill\n",
      "failure\n",
      "airport\n",
      "mile\n",
      "ft\n",
      "shortly\n",
      "\n",
      "\n",
      "Cluster 3\n",
      "mountain\n",
      "crash\n",
      "strike\n",
      "fly\n",
      "ft\n",
      "poor\n",
      "plane\n",
      "cargo\n",
      "fog\n",
      "approach\n",
      "weather\n",
      "course\n",
      "hit\n",
      "cloud\n",
      "aircraft\n",
      "visibility\n",
      "attempt\n",
      "mile\n",
      "shortly\n",
      "obscure\n",
      "\n",
      "\n",
      "Cluster 4\n",
      "weather\n",
      "condition\n",
      "vfr\n",
      "adverse\n",
      "poor\n",
      "flight\n",
      "continued\n",
      "pilot\n",
      "crash\n",
      "fly\n",
      "mountain\n",
      "ifr\n",
      "low\n",
      "terrain\n",
      "route\n",
      "attempt\n",
      "fog\n",
      "en\n",
      "altitude\n",
      "plane\n",
      "\n",
      "\n",
      "Cluster 5\n",
      "engine\n",
      "takeoff\n",
      "crash\n",
      "failure\n",
      "shortly\n",
      "plane\n",
      "lose\n",
      "aircraft\n",
      "emergency\n",
      "fail\n",
      "landing\n",
      "power\n",
      "right\n",
      "experience\n",
      "return\n",
      "cargo\n",
      "fuel\n",
      "left\n",
      "airport\n",
      "pilot\n"
     ]
    }
   ],
   "source": [
    "print('Most common terms')\n",
    "order_centroids = model.cluster_centers_.argsort()[:,::-1] #Sort cluster centers by proximity to centroid\n",
    "terms = tv.get_feature_names() #To get the word\n",
    "\n",
    "for i in range(6):\n",
    "    print('\\n')\n",
    "    print(f'Cluster {i}')\n",
    "    for j in order_centroids[i, :20]:\n",
    "        print(f'{terms[j]}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 912,
   "id": "74909abe",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2    2650\n",
       "1     874\n",
       "5     613\n",
       "4     276\n",
       "0     233\n",
       "3     232\n",
       "dtype: int64"
      ]
     },
     "execution_count": 912,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame(model.labels_).value_counts() #Most common cluster"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "06d19a7d",
   "metadata": {},
   "source": [
    "To test the clusters with words"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1196,
   "id": "1ec0dee7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([5])"
      ]
     },
     "execution_count": 1196,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "word = stem_punct_stop(nlp('engine failure'))\n",
    "\n",
    "model.predict(tv.transform([word]))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bbcaa88c",
   "metadata": {},
   "source": [
    "# Adding the cluster column to the dataframe and saving"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 923,
   "id": "c9576afc",
   "metadata": {},
   "outputs": [],
   "source": [
    "sum_index = df['Summary'].dropna().index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 924,
   "id": "3e242a07",
   "metadata": {},
   "outputs": [],
   "source": [
    "cluster_df = df[['Summary']].join(pd.DataFrame(model.labels_, index=sum_index, columns=['model']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 925,
   "id": "79d4628c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Summary</th>\n",
       "      <th>model</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>During a demonstration flight, a U.S. Army fly...</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>First U.S. dirigible Akron exploded just offsh...</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>The first fatal airplane accident in Canada oc...</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>The airship flew into a thunderstorm and encou...</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Hydrogen gas which was being vented was sucked...</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                             Summary  model\n",
       "0  During a demonstration flight, a U.S. Army fly...    2.0\n",
       "1  First U.S. dirigible Akron exploded just offsh...    2.0\n",
       "2  The first fatal airplane accident in Canada oc...    2.0\n",
       "3  The airship flew into a thunderstorm and encou...    2.0\n",
       "4  Hydrogen gas which was being vented was sucked...    2.0"
      ]
     },
     "execution_count": 925,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cluster_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 939,
   "id": "df1a8886",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Summary    390\n",
       "model      390\n",
       "dtype: int64"
      ]
     },
     "execution_count": 939,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cluster_df.isna().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 945,
   "id": "4d6fca2c",
   "metadata": {},
   "outputs": [],
   "source": [
    "df['cluster_label'] = cluster_df['model']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 947,
   "id": "1d824e6e",
   "metadata": {},
   "outputs": [],
   "source": [
    "fill_df = df.fillna('unknown')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 949,
   "id": "f1c84306",
   "metadata": {},
   "outputs": [],
   "source": [
    "fill_df.to_excel('complete.xlsx')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 950,
   "id": "020421a6",
   "metadata": {},
   "outputs": [],
   "source": [
    "df.to_excel('incomplete.xlsx')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "459aab65",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.8.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}