pr2, 95% complete

This commit is contained in:
ha4ska committed 2026-10-14 01:02:09 +05:00
1 parent 211e535766
commit d049732103
2 files changed
+7550

No files matched your search

File diff suppressed because it is too large. Load diff
+506
View File
@@ -0,0 +1,506 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "a7bd9ac6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(7043, 21)\n",
"<class 'pandas.DataFrame'>\n",
"RangeIndex: 7043 entries, 0 to 7042\n",
"Data columns (total 21 columns):\n",
" # Column Non-Null Count Dtype \n",
"--- ------ -------------- ----- \n",
" 0 customerID 7043 non-null str \n",
" 1 gender 7043 non-null str \n",
" 2 SeniorCitizen 7043 non-null int64 \n",
" 3 Partner 7043 non-null str \n",
" 4 Dependents 7043 non-null str \n",
" 5 tenure 7043 non-null int64 \n",
" 6 PhoneService 7043 non-null str \n",
" 7 MultipleLines 7043 non-null str \n",
" 8 InternetService 7043 non-null str \n",
" 9 OnlineSecurity 7043 non-null str \n",
" 10 OnlineBackup 7043 non-null str \n",
" 11 DeviceProtection 7043 non-null str \n",
" 12 TechSupport 7043 non-null str \n",
" 13 StreamingTV 7043 non-null str \n",
" 14 StreamingMovies 7043 non-null str \n",
" 15 Contract 7043 non-null str \n",
" 16 PaperlessBilling 7043 non-null str \n",
" 17 PaymentMethod 7043 non-null str \n",
" 18 MonthlyCharges 7043 non-null float64\n",
" 19 TotalCharges 7043 non-null str \n",
" 20 Churn 7043 non-null str \n",
"dtypes: float64(1), int64(2), str(18)\n",
"memory usage: 1.1 MB\n"
]
},
{
"data": {
"text/plain": [
"Churn\n",
"No 0.73463\n",
"Yes 0.26537\n",
"Name: proportion, dtype: float64"
]
},
"execution_count": 1,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import pandas as pd\n",
"import numpy as np\n",
"import matplotlib.pyplot as plt\n",
"import seaborn as sns\n",
"\n",
"df = pd.read_csv('Telco-Customer-Churn.csv')\n",
"print(df.shape)\n",
"df.head()\n",
"df.info()\n",
"df['Churn'].value_counts(normalize=True)"
]
},
{
"cell_type": "code",
"execution_count": 62,
"id": "f2826ebe",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"SeniorCitizen\n",
"0 0.837853\n",
"1 0.162147\n",
"Name: proportion, dtype: float64"
]
},
"execution_count": 62,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df['SeniorCitizen'].value_counts(normalize=True)"
]
},
{
"cell_type": "markdown",
"id": "145f387c",
"metadata": {},
"source": [
"1) Признаков 20, наблюдений 7043\n",
"2) Числовые признаки - SeniorCitizen, tenure, MonthlyCharges, все остальные категориальные\n",
"3) Пропусков нет\n",
"4) Перевес есть в сторону молодых (>83%)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "368304a6",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<Axes: >"
]
},
"execution_count": 2,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 640x480 with 2 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"sns.countplot(data=df, x=\"Churn\")\n",
"\n",
"sns.countplot(data=df, x=\"Contract\", hue='Churn')\n",
"\n",
"sns.boxplot(data=df, x='Churn', y='tenure')\n",
"\n",
"sns.heatmap(df.select_dtypes('number').corr(), annot=True, cmap='coolwarm')"
]
},
{
"cell_type": "markdown",
"id": "d07661e1",
"metadata": {},
"source": [
"Я считаю что самыми главными признаками оттока клиентов являются:\n",
"1) Стоимость в месяц (Mounthly Charges)\n",
"2) Общая стоимость (Total Charges)\n",
"3) Тех поддержка (Tech Support)\n",
"\n",
"Если про стоимость услуг все понятно, то вот техническая поддержка так же является признаком, потому что если при возникновении трудностей техническая поддержка не отработала должным образом то у клиента появляется желание перейти к другому оператору, а если такое происходит несколько раз то он и перейдет к другой компании."
]
},
{
"cell_type": "code",
"execution_count": 58,
"id": "8daed7c2",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<Axes: >"
]
},
"execution_count": 58,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "iVBORw0KGgoAAAANSUhEUgAAAoEAAAIUCAYAAABl+PZIAAAAOnRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjExLjIsIGh0dHBzOi8vbWF0cGxvdGxpYi5vcmcvgI3uAAAAAAlwSFlzAAAPYQAAD2EBqD+naQAAlvVJREFUeJzt3Qd4U2UXB/B/96KD0gll77333gIOZIj4MWQJgoCAMkQFVERRcA9ERRAFlaHsIXvvvUfZo3Tv3e85b02atGlooaXt7f/3PHna3NybprdpcnLe95zXIiUlJQVEREREVKhY5vUDICIiIqInj0EgERERUSHEIJCIiIioEGIQSERERFQIMQgkIiIiKoQYBBIREREVQgwCiYiIiAohBoFEREREhRCDQCIiIqJCiEEgERERUQ4KCwvD119/jR49euDXX3/N0jGhoaF477338OKLL2LcuHG4cOECchuDQCIiIqIcsn//flStWhVnz57FoUOHcObMmYceExUVhaZNm2LTpk3o1KkTgoODUb9+fZw8eRK5yTpX752IiIioEKlYsSIuXboEJycn7Ny5M0vHfP/997h3754KGosUKYLBgwfjxo0bePvtt7Fq1apce6zMBBIRERHlkGLFiqkAMDvWr1+Pzp07qwBQp3fv3iozmJSUhNzCIJCIiIjIjLi4OISHhxtdZFtO8ff3R8mSJY22yXX5GXfv3kVu4XAw5Zi1NpV5Nh9D2/kv8fw9opBmPXjuHtGbPxfluXsMk47z//ZR1d6QtaHS/PC+dGhqX8yYMcNo27Rp0zB9+vQcuX8J9hwdHY226bKCsbGxyC0MAomIiEiTLGwscuR+pkyZgvHjxxtts7OzQ05xdXVFSEiI0bagoCD1tWjR3PugxiCQiIiIyAwJ+HIy6Euvdu3aOHHihNG248ePw9fXV80xzC2cE0hERESaZGltkSOXnLZv3z50795dn+3r378/du/erbaLBw8eYOHChRgwYAByEzOBREREpEkWNk8+1xUaGoqXX35ZfS9tXlasWIHz58+jcuXK+Pjjj9X227dv459//lENpUWXLl3w5ptvokOHDmjQoIHqMSjZwXfeeSdXHyuDQCIiIqIc4uDgoA8CdV+Fu7u7/vtmzZph5cqV8PDw0G/76KOPMGLECJw7d04NA9epUwe5jUEgERERaVJuDOU+jMwdlKFec4oXL25ynzJlyqjLk8IgkIiIiDQpp6qDtYqFIURERESFEDOBREREpEl5MRxckDAIJCIiIk3icLB5HA4mIiIiKoSYCSQiIiJN4nCweQwCiYiISJMsrDgn0BwGgURERKRJlgwCzeKcQCIiIqJCiJlAIiIi0iQLSw4Hm8MgkIiIiDTJwooDnubw7BAREREVQswEEhERkSaxMMQ8BoFERESkSZwTaB6Hg4mIiIgKIWYCiYiISJM4HGweg0AiIiLSJK4YYh6Hg4mIiIgKIWYCiYiISJMsLJnrModBIBEREWkSq4PNYxBIREREmsTCEPOYJyUiIiIqhBgEZtOVK1cQGBiY7RN96dIlBAcHZ/s4IiIievTh4Jy4aJV1QQ3EYmJiMr3dzs4OFStWfOh9uLm5oVixYtn62X369EGvXr0wefLkDLeFhYUhICAAvr6+KFKkiNFtzzzzDF577TV1oSfL0t4OXt3awtLGGneWrin0pz8xORmHrwcgKCoGFTzdUNm7aJbOiex/9OYD2FtboUFpbzjYmH75uBsWhX3+d1HewxW1/Tw1d76v3biJS1f84erijDq1asDWxsbs/ikpKTh/6TLu3LkHd/eiqFmtCqytM567hIQEnDp7DqFh4ahepTK8vbR37gyV87NBcU8bBIUl4bx/HFJSzO/vYGeBCqVs4WhviZv3EnDnQWKGfaysgAolbVHM1QoPQpJw6UY8NMPCAk41asHWyxtxt24i+sK5hx9iZwen6rVg7eaGxKAgRJ4+ASQlGe1j6eAAx6o1YO3qhpirlxB3/Rq0hIUhGgwCp06ditOnT6vv4+LicPnyZZQrVw4ODg5qm3y/atUqs/fRs2dP9OvXD2+88cZjP57du3dj0qRJOHbsGEqUKIEHDx6or2+99Rb+97//Pfb906OrPHMC/Pp1R2J4JKzdnAt9EBgWE4cRS7YiMi4Blbzc8NGmw+hcrTTefqqR2fO46MA5fLfrJOqX9IKjrQ2+3nECn/VsheJuxh92EpKSMXHlblx6EIrutcppLgj8adHvWLFmHerVqokbt25D4pY5M6fBM5MPk/7Xb2DW3K9gZ2cLHy9PFTzGxMbio+lvo2zpkvr9Ll6+immzPoGTowPKlSmD3/5cgd7dn8FTHdpCaywsgJF93FGroh3O+cehnJ8tgkKT8NHPgYiLNx0Jdm1ZBE81L4I7AYmIiU3GK72K4uCpGMxbFqLfp1p5Owx6zg2hEUkIjUhG5TK2iI5JxqyfAhEWmYyCzNLOHmXfnw1b3+Iq+CtSszYijx/F9Y/fA5JN/24Olaqi3HsfI/7BfcTdvA6HSlVgYW2Dq5NeR/z9u2ofp1p1UHrKdMTfu4uEgPsoMWI0Qrb9izvff/mEf0PKKwUyCFy6dKn+ewkGa9asid9++w1NmjQx2u/u3buIiopC2bJlYSUfEf9z7do1xMbG4t69e/pgslq1arh16xbCw8NhYWEBLy8veHo+/A1s8+bN6Nq1q8oMyveOjo5q+8WLFzF//nyTQaA8LslWuru7G22/cePGQ3++DCtL9lKymNevX4eNjQ38/Pz0AbH8DiVLloStra363eR3d3JyMrqPO3fuICkpSe1nSmhoqLqUKlUKlgW8vD76ynXsqN0NpQb1Qtnxg1HYfbPjJOISk/DHkC4qmDt3Lxj9ftmIluWLo3XF1OdRelsv3MQX247juxfbolEZH7XtTmgkkkykbiQ4LOvhYvK2gu70ufP47a8VmPvhdNSpWQPxCQkYM3EqvvtxId6dNN7kMcnJyZg2eQJK+Pror78++V389Otv+ODt1NGEyKgoTH1/Fpo0rI/xo4ar///ExEScOX8RWtSiriMaVLPHlC8DcC8wEUUcLTFrrBeea+OMPzeFmzxGsnoT595H7H9BYgkva8wa443Tl2Ox53jqqFBiYgren/cA4VGpQZGtjQVmj/PG062d8dvaMBRkni+8pALAi6OGICk8DLbF/VD5259RtH1nhGxeb/IYnwGDVWbv6lsTUjdYWaHyd7+o+7r91Ry1qdSbbyNs7y79dRsPL1T+YREijh5CxMF90AItD+XmhIL9Dp8JCY6aNm2KChUqoEWLFihevDiWLVumv33mzJlqHwkcX3zxRXWRAOrjjz9W38uQb+XKlVG9enUcP37c7DDPq6++ik6dOuH999/XB4CiUqVK+OSTT4z2P3DggLrfZs2aqcfUu3dvFYzpZOXny7DymDFj1O/WoUMHfPPNN2r7ihUr4OPjo+7b29sbb775pgqODx06pD/2xIkTqFu3rtquewyrV6/W3x4dHY3nnntOZTHbt2+v7mfBggUoyG7+vAyJoabfWAobeb5uPHcdz9QspwJAUdXHHbX9PLDh7PVMj/tl/1m0qeSnDwCFZABLFnU22m/PlTvYduEmJnVsAC3aunMPypYupQJAIcPAT3fuiD37D6rXD1PKly2jDwCFfKgq5l4UiQlpQ5mbt+1EeEQkhr/cXwWAQoaLa9eoBi1qWtsRJy/FqQBQREYnY+/xGDSrnfb6md6h0zH6AFDcDkjEg5BE+HmnDcVfvB6vDwBFfEIKoqKTYW1V8IOAom06IHTHFhUAivg7txB+5KDanhn1YSI0LVMqw8BJEWmvhdZuRWHr6YXw/Xv02xICA1Tg6N6+M7RUHZwTF63SZBA4aNAgODs7q2FZyfZNmTIF/fv3V5k2IRk6CbImTJigsmVykaFkCah01+XYbt26qeMyc+rUKTW3cMiQIVl6XJs2bVLD1P7+/jh37hw2bNhgFJxm9eevW7dO3Y/87FmzZiEoKAgDBw5UQ9L3799XmUZ5bOnnKz711FNqPylsuXnzpvp5ffv2VZlR8dNPP6nHJcfLfcsw+6MUwVD+9CAyBuGx8Sjn4WK0XebuXXlgOlMSn5iEs3eD0aSMD64GhmHdaX8cvHZPbTcUGBmD6esO4P1nmsLJzvwcuYLq2vUbKF3SOFtaplRJJCQm4tad1OG1zGzZvgv/rNuIT7/8Tg0RD3u5n/6202fPoXKF8oAFsGPPPhVUBgZpt4ispLc1bt9PMNp2634CvIpZq+xdVvh6WMO7mDWu3TG+Hxm4aNPQEZ2aOuH1fsWQmJyCVdsL9odACxtblQWMTTdXL+66P+zLlM30uDs/fguH8hVR4tWxKPbM8yg18W11ggKWLFK3J4aHITk2Fg5ly6f9LFtb2BX3M3u/pC2aCwIvXLiAbdu24aOPPtJn5saOHauGTH/55ZeHHi/DNRIEnT9/Hp07d1YBmQRWpkg2UTcHMSskWJTgU8gwrQxfp8/0ZeXnDxgwADVqpGYjxJ9//qkKUST7J+zt7VW2M/0QumQhJBCU+5ZgTx6LFLFIcCoiIyPV0LEMVQtXV1f9fVLBJ/MAhYu9rdF2Vwc7RMYbv5nqRMTFq6HdHZduYdyyndh15Q5mbTyMnvPX4npw6ptrckoK3l69Dz3rlNfcHEBDUdExcC5iPLXC2Tl1TmR0dOaFauLilas4c+48Tp45C49i7vr5y0KygLFxsRg5YQq27dyjgsX+r7yGNRs2Q4sc7C0RFWM8j013XYo/HsbOxgKv9XVXRR8HTxufdxn5q1jKDpXL2KnCE5lraJB0LZAsHR1UcUNSZITR9sSIcFgajD6llxQdjYTAB6qYxKlqDThUrIz4+/dU4KckJ+Peop/g/b+XUfyVUfDo3hvlZ32mbrd0NH6eF2SsDjavQM4JNEeyV8IwSJK0uFzX3ZaZ33//XWUH4+Pj1Xw83Xy427dvq6HR9GTenW4YNStk+NWQBFwRERHZ/vkSQBqLine truncated
"text/plain": [
"<Figure size 640x480 with 2 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"df = pd.read_csv('Telco-Customer-Churn.csv')\n",
"\n",
"df[\"TotalCharges\"] = pd.to_numeric(df[\"TotalCharges\"], errors='coerce')\n",
"df[\"MonthlyCharges\"] = pd.to_numeric(df[\"MonthlyCharges\"], errors='coerce')\n",
"df[\"TechSupport_num\"] = df[\"TechSupport\"].map({'No':0, 'Yes':1}) \n",
"df[\"Churn_num\"] = df[\"Churn\"].map({'No':0, 'Yes':1}) \n",
"corr = df[[\"TotalCharges\", \"MonthlyCharges\", \"TechSupport_num\", \"Churn_num\", 'tenure']].corr(method='spearman')\n",
"\n",
"sns.heatmap(corr, annot=True, cmap='coolwarm')"
]
},
{
"cell_type": "markdown",
"id": "cd5e0ffc",
"metadata": {},
"source": [
"После того как я преобразовал числовые данные которые были поданы как строки в числа, и булевые состояния Нет и Да в 0 и 1 соответсвенно, можно наблюдать что общая стоимость +- -1\\4 влияния на отказ (меньше траты = больше удержание), а наличие тех поддержки почти -1\\3! (тех поддержка есть - уровень оттока ниже), в то время как оплата по месяцу чуть меньше 1\\5 влияния (дороже ежемесячный тариф = меньше удержание). Но есть еще одна метрика - tenure (Стаж клиента)? , которая имеет даже большее влияние на уход -0.37! (чем больше клиент с компанией тем меньше шанс что он уйдет)"
]
},
{
"cell_type": "code",
"execution_count": 44,
"id": "26350f63",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/tmp/ipykernel_9201/298299848.py:18: Pandas4Warning: For backward compatibility, 'str' dtypes are included by select_dtypes when 'object' dtype is specified. This behavior is deprecated and will be removed in a future version. Explicitly pass 'str' to `include` to select them, or to `exclude` to remove them and silence this warning.\n",
"See https://pandas.pydata.org/docs/user_guide/migration-3-strings.html#string-migration-select-dtypes for details on how to write code that works with pandas 2 and 3.\n",
" cat_cols = X.select_dtypes('object').columns.to_list()\n"
]
}
],
"source": [
"from sklearn.model_selection import train_test_split\n",
"from sklearn.preprocessing import StandardScaler, OneHotEncoder\n",
"from sklearn.compose import ColumnTransformer\n",
"from sklearn.pipeline import Pipeline\n",
"from sklearn.impute import SimpleImputer\n",
"\n",
"df = pd.read_csv('Telco-Customer-Churn.csv')\n",
"\n",
"df[\"Churn\"] = df[\"Churn\"].map({'No':0, 'Yes':1}) \n",
"df[\"TotalCharges\"] = pd.to_numeric(df[\"TotalCharges\"], errors='coerce')\n",
"\n",
"df = df.drop(columns=['customerID'])\n",
"\n",
"X = df.drop(columns=['Churn'])\n",
"y = df['Churn']\n",
"\n",
"num_cols = X.select_dtypes('number').columns.to_list()\n",
"cat_cols = X.select_dtypes('object').columns.to_list()\n",
"\n",
"preprocessor = ColumnTransformer([\n",
" ('num', Pipeline([\n",
" ('imputer', SimpleImputer(strategy='median')),\n",
" ('scaler', StandardScaler())\n",
" ]), num_cols),\n",
" ('cat', Pipeline([\n",
" ('imputer', SimpleImputer(strategy=\"most_frequent\")),\n",
" ('ohe', OneHotEncoder(handle_unknown='ignore'))\n",
" ]), cat_cols)\n",
"])\n",
"\n",
"X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, stratify=y, random_state=42)\n"
]
},
{
"cell_type": "markdown",
"id": "bbc50334",
"metadata": {},
"source": [
"stratify=y это параметр, который указывает что мы используем Стратифицированную выборку. При этой выборке мы делим популяцию на группы, после чего из каждой группы отбираем пропорциональные выборки. Т.е в нашем случае мы берем главый параметр (Churn) как метрику для пропорциональной выборки данных, т.е для каждой выборки из каждой группы будет справедливое распределение +- 73% тех кто остался и +- 27% тех кто ушел.\n",
"\n",
"hundle_unknown = \"ignore\" делается для того, что бы OneHotEncoder, который выполняет преобразование категориальных признаков в числовые массивы, когда встречает незнакомые категории преобразования, делает так что бы на выходе в столбце были 0. "
]
},
{
"cell_type": "code",
"execution_count": 46,
"id": "d5790ae5",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"=== LogReg ===\n",
" precision recall f1-score support\n",
"\n",
" 0 0.90 0.72 0.80 1035\n",
" 1 0.50 0.78 0.61 374\n",
"\n",
" accuracy 0.74 1409\n",
" macro avg 0.70 0.75 0.71 1409\n",
"weighted avg 0.80 0.74 0.75 1409\n",
"\n",
"ROC-AUC : 0.841\n",
"\n",
"=== RandomForest ===\n",
" precision recall f1-score support\n",
"\n",
" 0 0.86 0.82 0.84 1035\n",
" 1 0.55 0.62 0.59 374\n",
"\n",
" accuracy 0.77 1409\n",
" macro avg 0.70 0.72 0.71 1409\n",
"weighted avg 0.78 0.77 0.77 1409\n",
"\n",
"ROC-AUC : 0.822\n",
"\n",
"=== GradientBoosting ===\n",
" precision recall f1-score support\n",
"\n",
" 0 0.84 0.91 0.87 1035\n",
" 1 0.67 0.52 0.59 374\n",
"\n",
" accuracy 0.81 1409\n",
" macro avg 0.76 0.72 0.73 1409\n",
"weighted avg 0.80 0.81 0.80 1409\n",
"\n",
"ROC-AUC : 0.843\n",
"\n"
]
}
],
"source": [
"from sklearn.linear_model import LogisticRegression\n",
"from sklearn.ensemble import RandomForestClassifier\n",
"from sklearn.ensemble import GradientBoostingClassifier\n",
"from sklearn.metrics import (classification_report, roc_auc_score, confusion_matrix, RocCurveDisplay)\n",
"\n",
"models = {\n",
" 'LogReg': LogisticRegression(max_iter=100, class_weight='balanced'),\n",
" 'RandomForest' : RandomForestClassifier(n_estimators=300, class_weight='balanced', random_state=42),\n",
" 'GradientBoosting' : GradientBoostingClassifier(random_state=42)\n",
"}\n",
"\n",
"results = {}\n",
"\n",
"for name, clf in models.items():\n",
" pipe = Pipeline([('prep', preprocessor), ('clf', clf)])\n",
" pipe.fit(X_train, y_train)\n",
" y_pred = pipe.predict(X_val)\n",
" y_proba = pipe.predict_proba(X_val)[:, 1]\n",
" results[name] = {\n",
" 'roc_auc': roc_auc_score(y_val, y_proba),\n",
" 'report': classification_report(y_val, y_pred)\n",
" }\n",
"\n",
" print(f\"=== {name} ===\")\n",
" print(results[name]['report'])\n",
" print(f\"ROC-AUC : {results[name]['roc_auc']:.3f}\\n\")"
]
},
{
"cell_type": "markdown",
"id": "051a841e",
"metadata": {},
"source": [
"|Model | Accuracy | Precision (1) | Recall (1) | F1 | ROC-AUC |\n",
"| :---: | :---: | :---: | :---: | :---: | :---: |\n",
"| LogReg | 0.74 | 0.50 | 0.78 | 0.61 | 0.841 |\n",
"| RandForest | 0.77 | 0.55 | 0.62 | 0.59 | 0.822 |\n",
"| GradBoost | 0.81 | 0.67 | 0.52 | 0.59 | 0.843 |"
]
},
{
"cell_type": "code",
"execution_count": 48,
"id": "564ecece",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'clf__max_depth': 10, 'clf__min_samples_leaf': 5, 'clf__n_estimators': 400} 0.8477837278014084\n"
]
}
],
"source": [
"from sklearn.model_selection import GridSearchCV\n",
"\n",
"param_grid = {\n",
" 'clf__n_estimators': [200, 400],\n",
" 'clf__max_depth': [5, 10, None],\n",
" 'clf__min_samples_leaf': [1, 5] \n",
"}\n",
"\n",
"grid = GridSearchCV(\n",
" Pipeline([('prep', preprocessor), ('clf', RandomForestClassifier(random_state=42, class_weight='balanced'))]),\n",
" param_grid, cv=5, scoring='roc_auc', n_jobs=-1\n",
")\n",
"\n",
"grid.fit(X_train, y_train)\n",
"print(grid.best_params_, grid.best_score_)"
]
},
{
"cell_type": "markdown",
"id": "b5766d4f",
"metadata": {},
"source": [
"Качество улучшилось после подбора, но я бы не сказал что значительно, потому что лучший результат среди всех моделей показал GradientBooost 0.843, а после подбора стал 0.847"
]
},
{
"cell_type": "code",
"execution_count": 50,
"id": "54d0fe55",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<Axes: ylabel='Precision (Positive label: 1)'>"
]
},
"execution_count": 50,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 640x480 with 2 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 640x480 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"image/png": "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 truncated
"text/plain": [
"<Figure size 640x480 with 1 Axes>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"from sklearn.metrics import ConfusionMatrixDisplay, PrecisionRecallDisplay\n",
"\n",
"best = grid.best_estimator_\n",
"y_proba = best.predict_proba(X_val)[:,1]\n",
"y_pred = (y_proba >= 0.5).astype(int)\n",
"\n",
"ConfusionMatrixDisplay.from_predictions(y_val, y_pred)\n",
"RocCurveDisplay.from_predictions(y_val, y_proba)\n",
"PrecisionRecallDisplay.from_predictions(y_val, y_proba)\n",
"\n",
"ohe_cols = best.named_steps['prep'].named_transformers_['cat'].named_steps['ohe'].get_feature_names_out(cat_cols)\n",
"\n",
"feature_names = num_cols + list(ohe_cols)\n",
"importances = pd.Series(best.named_steps['clf'].feature_importances_, index=feature_names)\n",
"importances.sort_values(ascending=False).head(15).plot(kind='barh')"
]
},
{
"cell_type": "markdown",
"id": "6c1672fe",
"metadata": {},
"source": [
"Для любой компании False Negative в разы хуже чем False Positive, так как при наличии проблемы они не смогут ее устранить и будут терять деньги на этом"
]
},
{
"cell_type": "markdown",
"id": "86f87c17",
"metadata": {},
"source": [
"Можно не подбирать порог классификации заново, а обратиться к Logical Regression - там как раз recall = 0.78, и мы видим что precision при этом падает, соответсвенно precision обратно пропорционален recall"
]
},
{
"cell_type": "markdown",
"id": "4ae24fff",
"metadata": {},
"source": [
"## Формулирование бизнес выводов.\n",
"1. Цена в месяц, стаж клиента, наличие тех поддержки и общая стоимость\n",
"2. Gradient Boost, так как она использует алгоритм градиентного спуска при оптимизации, что позволяет создать мощную предсказательную модель, потому что градиентный спуск двигает веса в сторону наименьшего значения функции потерь, а градинетный буст добавляет пошагово простые модели которые исправляют ошибки прошлых (берут значения прошлых деревьев решений и исправляет их ошибки)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "ai_lesson",
"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": "3.12.15"
}
},
"nbformat": 4,
"nbformat_minor": 5
}