Spaces:
Sleeping
Sleeping
thanthamky
commited on
Commit
•
914e01f
1
Parent(s):
99e0926
Upload 3 files
Browse files- app/1-eda.ipynb +0 -0
- app/2-data_preprocessing.ipynb +326 -0
- app/3-modeling.ipynb +830 -0
app/1-eda.ipynb
ADDED
The diff for this file is too large to render.
See raw diff
|
|
app/2-data_preprocessing.ipynb
ADDED
@@ -0,0 +1,326 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cells": [
|
3 |
+
{
|
4 |
+
"attachments": {},
|
5 |
+
"cell_type": "markdown",
|
6 |
+
"metadata": {},
|
7 |
+
"source": [
|
8 |
+
"# Data Preprocessing\n",
|
9 |
+
"\n",
|
10 |
+
"This file shows how I performed data cleaning and feature engineering. "
|
11 |
+
]
|
12 |
+
},
|
13 |
+
{
|
14 |
+
"cell_type": "markdown",
|
15 |
+
"metadata": {},
|
16 |
+
"source": [
|
17 |
+
"## Set up"
|
18 |
+
]
|
19 |
+
},
|
20 |
+
{
|
21 |
+
"attachments": {},
|
22 |
+
"cell_type": "markdown",
|
23 |
+
"metadata": {},
|
24 |
+
"source": [
|
25 |
+
"Import libraries."
|
26 |
+
]
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"cell_type": "code",
|
30 |
+
"execution_count": 2,
|
31 |
+
"metadata": {},
|
32 |
+
"outputs": [],
|
33 |
+
"source": [
|
34 |
+
"import numpy as np\n",
|
35 |
+
"import pandas as pd\n",
|
36 |
+
"from sklearn.model_selection import train_test_split\n",
|
37 |
+
"from sklearn.preprocessing import MinMaxScaler"
|
38 |
+
]
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"cell_type": "code",
|
42 |
+
"execution_count": null,
|
43 |
+
"metadata": {},
|
44 |
+
"outputs": [],
|
45 |
+
"source": [
|
46 |
+
"#!pip install scikit-learn"
|
47 |
+
]
|
48 |
+
},
|
49 |
+
{
|
50 |
+
"cell_type": "markdown",
|
51 |
+
"metadata": {},
|
52 |
+
"source": [
|
53 |
+
"Load datasets."
|
54 |
+
]
|
55 |
+
},
|
56 |
+
{
|
57 |
+
"cell_type": "code",
|
58 |
+
"execution_count": 3,
|
59 |
+
"metadata": {},
|
60 |
+
"outputs": [],
|
61 |
+
"source": [
|
62 |
+
"df_train_full = pd.read_csv(\"https://raw.githubusercontent.com/kingyiusuen/travelers-insurance-fraud/master/data/raw/train.csv\")\n",
|
63 |
+
"df_test = pd.read_csv(\"https://raw.githubusercontent.com/kingyiusuen/travelers-insurance-fraud/master/data/raw/test.csv\")"
|
64 |
+
]
|
65 |
+
},
|
66 |
+
{
|
67 |
+
"attachments": {},
|
68 |
+
"cell_type": "markdown",
|
69 |
+
"metadata": {},
|
70 |
+
"source": [
|
71 |
+
"Since the test set provided does not have the target variable, so we have to create an internal validation set to evaluate the model performance."
|
72 |
+
]
|
73 |
+
},
|
74 |
+
{
|
75 |
+
"cell_type": "code",
|
76 |
+
"execution_count": 4,
|
77 |
+
"metadata": {},
|
78 |
+
"outputs": [],
|
79 |
+
"source": [
|
80 |
+
"df_train, df_val = train_test_split(df_train_full, test_size=0.2, random_state=99)"
|
81 |
+
]
|
82 |
+
},
|
83 |
+
{
|
84 |
+
"cell_type": "markdown",
|
85 |
+
"metadata": {},
|
86 |
+
"source": [
|
87 |
+
"## Data Cleaning"
|
88 |
+
]
|
89 |
+
},
|
90 |
+
{
|
91 |
+
"cell_type": "markdown",
|
92 |
+
"metadata": {},
|
93 |
+
"source": [
|
94 |
+
"Remove the observations whose the target variable `fraud` is equal to -1."
|
95 |
+
]
|
96 |
+
},
|
97 |
+
{
|
98 |
+
"cell_type": "code",
|
99 |
+
"execution_count": 5,
|
100 |
+
"metadata": {},
|
101 |
+
"outputs": [],
|
102 |
+
"source": [
|
103 |
+
"df_train = df_train[df_train[\"fraud\"] != -1]\n",
|
104 |
+
"df_val = df_val[df_val[\"fraud\"] != -1]"
|
105 |
+
]
|
106 |
+
},
|
107 |
+
{
|
108 |
+
"cell_type": "markdown",
|
109 |
+
"metadata": {},
|
110 |
+
"source": [
|
111 |
+
"For values that match the following conditions, treat them as missing values to be imputed later.\n",
|
112 |
+
"\n",
|
113 |
+
"- `age_of_driver > 100`\n",
|
114 |
+
"- `annual_income = -1`\n",
|
115 |
+
"- `zip_code = -1`\n",
|
116 |
+
"\n",
|
117 |
+
"According to [Wikipedia](https://en.wikipedia.org/wiki/List_of_the_verified_oldest_people), the oldest living person is 115, as of 2018. I think it is reasonable to assume that any `age_of_driver > 100` in this dataset is a clerical error."
|
118 |
+
]
|
119 |
+
},
|
120 |
+
{
|
121 |
+
"cell_type": "code",
|
122 |
+
"execution_count": 6,
|
123 |
+
"metadata": {},
|
124 |
+
"outputs": [],
|
125 |
+
"source": [
|
126 |
+
"for df in [df_train, df_val, df_test]:\n",
|
127 |
+
" df.loc[df[\"age_of_driver\"] > 100, \"age_of_driver\"] = np.nan\n",
|
128 |
+
" df.loc[df[\"annual_income\"] == -1, \"annual_income\"] = np.nan\n",
|
129 |
+
" df.loc[df[\"zip_code\"] == 0, \"zip_code\"] = np.nan"
|
130 |
+
]
|
131 |
+
},
|
132 |
+
{
|
133 |
+
"attachments": {},
|
134 |
+
"cell_type": "markdown",
|
135 |
+
"metadata": {},
|
136 |
+
"source": [
|
137 |
+
"Now, we will do an imputation for the missing values. Since there is only a very small percentage of missing values, we will simply do a mean/mode imputation for the continuous/categorical variables. Note that the mean/mode is computed based on the training set only to prevent data leakage."
|
138 |
+
]
|
139 |
+
},
|
140 |
+
{
|
141 |
+
"cell_type": "code",
|
142 |
+
"execution_count": 7,
|
143 |
+
"metadata": {},
|
144 |
+
"outputs": [
|
145 |
+
{
|
146 |
+
"name": "stderr",
|
147 |
+
"output_type": "stream",
|
148 |
+
"text": [
|
149 |
+
"/tmp/ipykernel_293/883070373.py:5: FutureWarning: A value is trying to be set on a copy of a DataFrame or Series through chained assignment using an inplace method.\n",
|
150 |
+
"The behavior will change in pandas 3.0. This inplace method will never work because the intermediate object on which we are setting values always behaves as a copy.\n",
|
151 |
+
"\n",
|
152 |
+
"For example, when doing 'df[col].method(value, inplace=True)', try using 'df.method({col: value}, inplace=True)' or df[col] = df[col].method(value) instead, to perform the operation inplace on the original object.\n",
|
153 |
+
"\n",
|
154 |
+
"\n",
|
155 |
+
" df[feature].fillna(int(feature_mean), inplace=True)\n",
|
156 |
+
"/tmp/ipykernel_293/883070373.py:10: FutureWarning: A value is trying to be set on a copy of a DataFrame or Series through chained assignment using an inplace method.\n",
|
157 |
+
"The behavior will change in pandas 3.0. This inplace method will never work because the intermediate object on which we are setting values always behaves as a copy.\n",
|
158 |
+
"\n",
|
159 |
+
"For example, when doing 'df[col].method(value, inplace=True)', try using 'df.method({col: value}, inplace=True)' or df[col] = df[col].method(value) instead, to perform the operation inplace on the original object.\n",
|
160 |
+
"\n",
|
161 |
+
"\n",
|
162 |
+
" df[feature].fillna(feature_mode.values[0], inplace=True)\n"
|
163 |
+
]
|
164 |
+
}
|
165 |
+
],
|
166 |
+
"source": [
|
167 |
+
"for df in [df_train, df_val, df_test]:\n",
|
168 |
+
" # mean imputation for continuous variables\n",
|
169 |
+
" for feature in [\"age_of_driver\", \"annual_income\", \"claim_est_payout\", \"age_of_vehicle\"]:\n",
|
170 |
+
" feature_mean = df_train.loc[:, feature].mean(skipna=True)\n",
|
171 |
+
" df[feature].fillna(int(feature_mean), inplace=True)\n",
|
172 |
+
"\n",
|
173 |
+
" # mode imputation for categorical variables\n",
|
174 |
+
" for feature in [\"marital_status\", \"witness_present_ind\", \"zip_code\"]:\n",
|
175 |
+
" feature_mode = df_train.loc[:, feature].mode(dropna=True)\n",
|
176 |
+
" df[feature].fillna(feature_mode.values[0], inplace=True)"
|
177 |
+
]
|
178 |
+
},
|
179 |
+
{
|
180 |
+
"cell_type": "markdown",
|
181 |
+
"metadata": {},
|
182 |
+
"source": [
|
183 |
+
"## Feature Engineering"
|
184 |
+
]
|
185 |
+
},
|
186 |
+
{
|
187 |
+
"attachments": {},
|
188 |
+
"cell_type": "markdown",
|
189 |
+
"metadata": {},
|
190 |
+
"source": [
|
191 |
+
"Remove features that do not seem to be related to the target variable (based on common sense)."
|
192 |
+
]
|
193 |
+
},
|
194 |
+
{
|
195 |
+
"cell_type": "code",
|
196 |
+
"execution_count": 8,
|
197 |
+
"metadata": {},
|
198 |
+
"outputs": [],
|
199 |
+
"source": [
|
200 |
+
"for df in [df_train, df_val, df_test]:\n",
|
201 |
+
" df.drop(columns=[\"claim_date\", \"claim_day_of_week\", \"vehicle_color\"], inplace=True)"
|
202 |
+
]
|
203 |
+
},
|
204 |
+
{
|
205 |
+
"attachments": {},
|
206 |
+
"cell_type": "markdown",
|
207 |
+
"metadata": {},
|
208 |
+
"source": [
|
209 |
+
"There are many unique `zip_code`. Creating dummy variables for `zip_code` will increase the dimensionality of the data too much. One idea is to transform it into `latitude` and `longitude` using the data from [UnitedStatesZipCodes.org](https://www.unitedstateszipcodes.org/zip-code-database/)."
|
210 |
+
]
|
211 |
+
},
|
212 |
+
{
|
213 |
+
"cell_type": "code",
|
214 |
+
"execution_count": 10,
|
215 |
+
"metadata": {},
|
216 |
+
"outputs": [],
|
217 |
+
"source": [
|
218 |
+
"zip_code_database = pd.read_csv(\"https://raw.githubusercontent.com/kingyiusuen/travelers-insurance-fraud/master/data/external/zip_code_database.csv\")\n",
|
219 |
+
"latitude_and_longitude_lookup = {\n",
|
220 |
+
" row.zip: (row.latitude, row.longitude) for row in zip_code_database.itertuples()\n",
|
221 |
+
"}\n",
|
222 |
+
"\n",
|
223 |
+
"for df in [df_train, df_val, df_test]:\n",
|
224 |
+
" df[\"latitude\"] = df[\"zip_code\"].apply(lambda x: latitude_and_longitude_lookup[x][0])\n",
|
225 |
+
" df[\"longitude\"] = df[\"zip_code\"].apply(lambda x: latitude_and_longitude_lookup[x][1])"
|
226 |
+
]
|
227 |
+
},
|
228 |
+
{
|
229 |
+
"attachments": {},
|
230 |
+
"cell_type": "markdown",
|
231 |
+
"metadata": {},
|
232 |
+
"source": [
|
233 |
+
"Another idea is to use [target encoding](https://maxhalford.github.io/blog/target-encoding/), but after a few experiments it seems to perform worse than just transforming it to `latitude` and `longitude`."
|
234 |
+
]
|
235 |
+
},
|
236 |
+
{
|
237 |
+
"cell_type": "code",
|
238 |
+
"execution_count": 11,
|
239 |
+
"metadata": {},
|
240 |
+
"outputs": [],
|
241 |
+
"source": [
|
242 |
+
"#from category_encoders.target_encoder import TargetEncoder\n",
|
243 |
+
"#\n",
|
244 |
+
"#target_encoder = TargetEncoder(cols=[\"zip_code\"], smoothing=10)\n",
|
245 |
+
"#target_encoder.fit(df_train[\"zip_code\"], df_train[\"fraud\"])\n",
|
246 |
+
"#\n",
|
247 |
+
"#for df in [df_train, df_val, df_test]:\n",
|
248 |
+
"# df[\"zip_code_target_encoded\"] = target_encoder.transform(df[\"zip_code\"])"
|
249 |
+
]
|
250 |
+
},
|
251 |
+
{
|
252 |
+
"attachments": {},
|
253 |
+
"cell_type": "markdown",
|
254 |
+
"metadata": {},
|
255 |
+
"source": [
|
256 |
+
"Now we can drop `zip_code`."
|
257 |
+
]
|
258 |
+
},
|
259 |
+
{
|
260 |
+
"cell_type": "code",
|
261 |
+
"execution_count": 12,
|
262 |
+
"metadata": {},
|
263 |
+
"outputs": [],
|
264 |
+
"source": [
|
265 |
+
"for df in [df_train, df_val, df_test]:\n",
|
266 |
+
" df.drop(columns=[\"zip_code\"], inplace=True)"
|
267 |
+
]
|
268 |
+
},
|
269 |
+
{
|
270 |
+
"cell_type": "markdown",
|
271 |
+
"metadata": {},
|
272 |
+
"source": [
|
273 |
+
"## Export processed data"
|
274 |
+
]
|
275 |
+
},
|
276 |
+
{
|
277 |
+
"cell_type": "code",
|
278 |
+
"execution_count": 14,
|
279 |
+
"metadata": {},
|
280 |
+
"outputs": [],
|
281 |
+
"source": [
|
282 |
+
"#df_train.to_csv(\"../data/processed/train.csv\", index=False)\n",
|
283 |
+
"#df_val.to_csv(\"../data/processed/val.csv\", index=False)\n",
|
284 |
+
"#df_test.to_csv(\"../data/processed/test.csv\", index=False)"
|
285 |
+
]
|
286 |
+
},
|
287 |
+
{
|
288 |
+
"cell_type": "code",
|
289 |
+
"execution_count": null,
|
290 |
+
"metadata": {},
|
291 |
+
"outputs": [],
|
292 |
+
"source": []
|
293 |
+
},
|
294 |
+
{
|
295 |
+
"cell_type": "code",
|
296 |
+
"execution_count": null,
|
297 |
+
"metadata": {},
|
298 |
+
"outputs": [],
|
299 |
+
"source": []
|
300 |
+
}
|
301 |
+
],
|
302 |
+
"metadata": {
|
303 |
+
"interpreter": {
|
304 |
+
"hash": "03e93f2959c516196957ae17ec0aa5d1e9fc5dd82cbe13968d4cfc2a60558992"
|
305 |
+
},
|
306 |
+
"kernelspec": {
|
307 |
+
"display_name": "Python 3 (ipykernel)",
|
308 |
+
"language": "python",
|
309 |
+
"name": "python3"
|
310 |
+
},
|
311 |
+
"language_info": {
|
312 |
+
"codemirror_mode": {
|
313 |
+
"name": "ipython",
|
314 |
+
"version": 3
|
315 |
+
},
|
316 |
+
"file_extension": ".py",
|
317 |
+
"mimetype": "text/x-python",
|
318 |
+
"name": "python",
|
319 |
+
"nbconvert_exporter": "python",
|
320 |
+
"pygments_lexer": "ipython3",
|
321 |
+
"version": "3.12.1"
|
322 |
+
}
|
323 |
+
},
|
324 |
+
"nbformat": 4,
|
325 |
+
"nbformat_minor": 4
|
326 |
+
}
|
app/3-modeling.ipynb
ADDED
@@ -0,0 +1,830 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cells": [
|
3 |
+
{
|
4 |
+
"attachments": {},
|
5 |
+
"cell_type": "markdown",
|
6 |
+
"metadata": {},
|
7 |
+
"source": [
|
8 |
+
"# Modeling\n",
|
9 |
+
"\n",
|
10 |
+
"In this notebook, the performance of different models is examined."
|
11 |
+
]
|
12 |
+
},
|
13 |
+
{
|
14 |
+
"cell_type": "markdown",
|
15 |
+
"metadata": {},
|
16 |
+
"source": [
|
17 |
+
"## Setup"
|
18 |
+
]
|
19 |
+
},
|
20 |
+
{
|
21 |
+
"attachments": {},
|
22 |
+
"cell_type": "markdown",
|
23 |
+
"metadata": {},
|
24 |
+
"source": [
|
25 |
+
"Import libraries."
|
26 |
+
]
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"cell_type": "code",
|
30 |
+
"execution_count": 35,
|
31 |
+
"metadata": {},
|
32 |
+
"outputs": [],
|
33 |
+
"source": [
|
34 |
+
"import numpy as np\n",
|
35 |
+
"import pandas as pd\n",
|
36 |
+
"from imblearn.pipeline import make_pipeline\n",
|
37 |
+
"from imblearn.over_sampling import SMOTE\n",
|
38 |
+
"from sklearn.compose import make_column_transformer\n",
|
39 |
+
"from sklearn.ensemble import GradientBoostingClassifier, RandomForestClassifier\n",
|
40 |
+
"from sklearn.linear_model import LogisticRegression\n",
|
41 |
+
"from sklearn.metrics import roc_auc_score\n",
|
42 |
+
"from sklearn.model_selection import cross_val_score, GridSearchCV, RandomizedSearchCV\n",
|
43 |
+
"from sklearn.neighbors import KNeighborsClassifier\n",
|
44 |
+
"from sklearn.preprocessing import OneHotEncoder, MinMaxScaler\n",
|
45 |
+
"from xgboost import XGBClassifier"
|
46 |
+
]
|
47 |
+
},
|
48 |
+
{
|
49 |
+
"cell_type": "code",
|
50 |
+
"execution_count": 5,
|
51 |
+
"metadata": {},
|
52 |
+
"outputs": [
|
53 |
+
{
|
54 |
+
"name": "stdout",
|
55 |
+
"output_type": "stream",
|
56 |
+
"text": [
|
57 |
+
"Requirement already satisfied: imblearn in /home/user/miniconda/lib/python3.12/site-packages (0.0)\n",
|
58 |
+
"Collecting xgboost\n",
|
59 |
+
" Downloading xgboost-2.0.3-py3-none-manylinux2014_x86_64.whl.metadata (2.0 kB)\n",
|
60 |
+
"Requirement already satisfied: imbalanced-learn in /home/user/miniconda/lib/python3.12/site-packages (from imblearn) (0.12.3)\n",
|
61 |
+
"Requirement already satisfied: numpy in /home/user/miniconda/lib/python3.12/site-packages (from xgboost) (1.26.4)\n",
|
62 |
+
"Requirement already satisfied: scipy in /home/user/miniconda/lib/python3.12/site-packages (from xgboost) (1.13.1)\n",
|
63 |
+
"Requirement already satisfied: scikit-learn>=1.0.2 in /home/user/miniconda/lib/python3.12/site-packages (from imbalanced-learn->imblearn) (1.5.0)\n",
|
64 |
+
"Requirement already satisfied: joblib>=1.1.1 in /home/user/miniconda/lib/python3.12/site-packages (from imbalanced-learn->imblearn) (1.4.2)\n",
|
65 |
+
"Requirement already satisfied: threadpoolctl>=2.0.0 in /home/user/miniconda/lib/python3.12/site-packages (from imbalanced-learn->imblearn) (3.5.0)\n",
|
66 |
+
"Downloading xgboost-2.0.3-py3-none-manylinux2014_x86_64.whl (297.1 MB)\n",
|
67 |
+
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m297.1/297.1 MB\u001b[0m \u001b[31m3.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m:00:01\u001b[0m00:01\u001b[0m\n",
|
68 |
+
"\u001b[?25hInstalling collected packages: xgboost\n",
|
69 |
+
"Successfully installed xgboost-2.0.3\n"
|
70 |
+
]
|
71 |
+
}
|
72 |
+
],
|
73 |
+
"source": [
|
74 |
+
"!pip install imblearn xgboost"
|
75 |
+
]
|
76 |
+
},
|
77 |
+
{
|
78 |
+
"cell_type": "markdown",
|
79 |
+
"metadata": {},
|
80 |
+
"source": [
|
81 |
+
"Load datasets."
|
82 |
+
]
|
83 |
+
},
|
84 |
+
{
|
85 |
+
"cell_type": "code",
|
86 |
+
"execution_count": 36,
|
87 |
+
"metadata": {},
|
88 |
+
"outputs": [],
|
89 |
+
"source": [
|
90 |
+
"df_train = pd.read_csv(\"https://raw.githubusercontent.com/kingyiusuen/travelers-insurance-fraud/master/data/processed/train.csv\")\n",
|
91 |
+
"df_val = pd.read_csv(\"https://raw.githubusercontent.com/kingyiusuen/travelers-insurance-fraud/master/data/processed/val.csv\")\n",
|
92 |
+
"df_test = pd.read_csv(\"https://raw.githubusercontent.com/kingyiusuen/travelers-insurance-fraud/master/data/processed/test.csv\")"
|
93 |
+
]
|
94 |
+
},
|
95 |
+
{
|
96 |
+
"cell_type": "code",
|
97 |
+
"execution_count": 37,
|
98 |
+
"metadata": {},
|
99 |
+
"outputs": [],
|
100 |
+
"source": [
|
101 |
+
"X_train = df_train.drop(columns=[\"claim_number\", \"fraud\"])\n",
|
102 |
+
"y_train = df_train[\"fraud\"]\n",
|
103 |
+
"X_val = df_val.drop(columns=[\"claim_number\", \"fraud\"])\n",
|
104 |
+
"y_val = df_val[\"fraud\"]\n",
|
105 |
+
"X_test = df_test.drop(columns=[\"claim_number\"])"
|
106 |
+
]
|
107 |
+
},
|
108 |
+
{
|
109 |
+
"cell_type": "markdown",
|
110 |
+
"metadata": {},
|
111 |
+
"source": [
|
112 |
+
"## Model Selection"
|
113 |
+
]
|
114 |
+
},
|
115 |
+
{
|
116 |
+
"attachments": {},
|
117 |
+
"cell_type": "markdown",
|
118 |
+
"metadata": {},
|
119 |
+
"source": [
|
120 |
+
"`OneHotEncoder` will dummify categorical features, and numerical features will be re-scaled with `MinMaxScaler`."
|
121 |
+
]
|
122 |
+
},
|
123 |
+
{
|
124 |
+
"cell_type": "code",
|
125 |
+
"execution_count": 38,
|
126 |
+
"metadata": {},
|
127 |
+
"outputs": [],
|
128 |
+
"source": [
|
129 |
+
"categorical_features = X_train.columns[X_train.dtypes == object].tolist()\n",
|
130 |
+
"column_transformer = make_column_transformer(\n",
|
131 |
+
" (OneHotEncoder(drop=\"first\"), categorical_features),\n",
|
132 |
+
" remainder=\"passthrough\",\n",
|
133 |
+
")\n",
|
134 |
+
"scaler = MinMaxScaler()"
|
135 |
+
]
|
136 |
+
},
|
137 |
+
{
|
138 |
+
"attachments": {},
|
139 |
+
"cell_type": "markdown",
|
140 |
+
"metadata": {},
|
141 |
+
"source": [
|
142 |
+
"A simple function that defines the training pipeline: fit the model, predict on the validation set, print the evaluation metric."
|
143 |
+
]
|
144 |
+
},
|
145 |
+
{
|
146 |
+
"cell_type": "code",
|
147 |
+
"execution_count": 39,
|
148 |
+
"metadata": {},
|
149 |
+
"outputs": [],
|
150 |
+
"source": [
|
151 |
+
"def modeling(X_train, y_train, X_val, y_val, steps):\n",
|
152 |
+
" pipeline = make_pipeline(*steps)\n",
|
153 |
+
" pipeline.fit(X_train, y_train)\n",
|
154 |
+
" y_val_pred = pipeline.predict_proba(X_val)[:, 1]\n",
|
155 |
+
" metric = roc_auc_score(y_val, y_val_pred)\n",
|
156 |
+
" if isinstance(pipeline._final_estimator, RandomizedSearchCV) or isinstance(pipeline._final_estimator, GridSearchCV):\n",
|
157 |
+
" print(f\"Best params: {pipeline._final_estimator.best_params_}\")\n",
|
158 |
+
" print(f\"AUC score: {metric}\")\n",
|
159 |
+
" return pipeline"
|
160 |
+
]
|
161 |
+
},
|
162 |
+
{
|
163 |
+
"attachments": {},
|
164 |
+
"cell_type": "markdown",
|
165 |
+
"metadata": {},
|
166 |
+
"source": [
|
167 |
+
"### K-Nearest Neighbor\n",
|
168 |
+
"\n",
|
169 |
+
"KNN has two hyperparameters: the number of neighbors, and whether all points in each neighborhood are weighted equally or weighted by the inverse of their distance. Since the number of hyperparameters is small. A grid search is used to find the optimal hyperparameter values."
|
170 |
+
]
|
171 |
+
},
|
172 |
+
{
|
173 |
+
"cell_type": "code",
|
174 |
+
"execution_count": 40,
|
175 |
+
"metadata": {},
|
176 |
+
"outputs": [
|
177 |
+
{
|
178 |
+
"name": "stdout",
|
179 |
+
"output_type": "stream",
|
180 |
+
"text": [
|
181 |
+
"Best params: {'n_neighbors': 50, 'weights': 'distance'}\n",
|
182 |
+
"AUC score: 0.6507841602442943\n"
|
183 |
+
]
|
184 |
+
}
|
185 |
+
],
|
186 |
+
"source": [
|
187 |
+
"param_grid = {\n",
|
188 |
+
" \"n_neighbors\": [5, 10, 25, 50],\n",
|
189 |
+
" \"weights\": [\"uniform\", \"distance\"],\n",
|
190 |
+
"}\n",
|
191 |
+
"\n",
|
192 |
+
"knn_clf = GridSearchCV(\n",
|
193 |
+
" KNeighborsClassifier(),\n",
|
194 |
+
" param_grid=param_grid,\n",
|
195 |
+
" n_jobs=-1,\n",
|
196 |
+
" cv=5,\n",
|
197 |
+
" scoring=\"roc_auc\",\n",
|
198 |
+
")\n",
|
199 |
+
"\n",
|
200 |
+
"knn_pipeline = modeling(X_train, y_train, X_val, y_val, [column_transformer, scaler, knn_clf])"
|
201 |
+
]
|
202 |
+
},
|
203 |
+
{
|
204 |
+
"attachments": {},
|
205 |
+
"cell_type": "markdown",
|
206 |
+
"metadata": {},
|
207 |
+
"source": [
|
208 |
+
"### Logistic Regression\n",
|
209 |
+
"\n",
|
210 |
+
"For logistic regression, there is no hyperparameter to tune."
|
211 |
+
]
|
212 |
+
},
|
213 |
+
{
|
214 |
+
"cell_type": "code",
|
215 |
+
"execution_count": 41,
|
216 |
+
"metadata": {},
|
217 |
+
"outputs": [
|
218 |
+
{
|
219 |
+
"name": "stdout",
|
220 |
+
"output_type": "stream",
|
221 |
+
"text": [
|
222 |
+
"AUC score: 0.7157014847720347\n"
|
223 |
+
]
|
224 |
+
}
|
225 |
+
],
|
226 |
+
"source": [
|
227 |
+
"lr_clf = LogisticRegression()\n",
|
228 |
+
"lr_pipeline = modeling(X_train, y_train, X_val, y_val, [column_transformer, scaler, lr_clf])"
|
229 |
+
]
|
230 |
+
},
|
231 |
+
{
|
232 |
+
"attachments": {},
|
233 |
+
"cell_type": "markdown",
|
234 |
+
"metadata": {},
|
235 |
+
"source": [
|
236 |
+
"Look at the model coefficients."
|
237 |
+
]
|
238 |
+
},
|
239 |
+
{
|
240 |
+
"cell_type": "code",
|
241 |
+
"execution_count": 42,
|
242 |
+
"metadata": {},
|
243 |
+
"outputs": [
|
244 |
+
{
|
245 |
+
"data": {
|
246 |
+
"text/html": [
|
247 |
+
"<div>\n",
|
248 |
+
"<style scoped>\n",
|
249 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
250 |
+
" vertical-align: middle;\n",
|
251 |
+
" }\n",
|
252 |
+
"\n",
|
253 |
+
" .dataframe tbody tr th {\n",
|
254 |
+
" vertical-align: top;\n",
|
255 |
+
" }\n",
|
256 |
+
"\n",
|
257 |
+
" .dataframe thead th {\n",
|
258 |
+
" text-align: right;\n",
|
259 |
+
" }\n",
|
260 |
+
"</style>\n",
|
261 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
262 |
+
" <thead>\n",
|
263 |
+
" <tr style=\"text-align: right;\">\n",
|
264 |
+
" <th></th>\n",
|
265 |
+
" <th>feature_name</th>\n",
|
266 |
+
" <th>coefficient</th>\n",
|
267 |
+
" </tr>\n",
|
268 |
+
" </thead>\n",
|
269 |
+
" <tbody>\n",
|
270 |
+
" <tr>\n",
|
271 |
+
" <th>0</th>\n",
|
272 |
+
" <td>past_num_of_claims</td>\n",
|
273 |
+
" <td>1.750160</td>\n",
|
274 |
+
" </tr>\n",
|
275 |
+
" <tr>\n",
|
276 |
+
" <th>1</th>\n",
|
277 |
+
" <td>annual_income</td>\n",
|
278 |
+
" <td>1.570769</td>\n",
|
279 |
+
" </tr>\n",
|
280 |
+
" <tr>\n",
|
281 |
+
" <th>2</th>\n",
|
282 |
+
" <td>age_of_vehicle</td>\n",
|
283 |
+
" <td>0.982407</td>\n",
|
284 |
+
" </tr>\n",
|
285 |
+
" <tr>\n",
|
286 |
+
" <th>3</th>\n",
|
287 |
+
" <td>address_change_ind</td>\n",
|
288 |
+
" <td>0.398596</td>\n",
|
289 |
+
" </tr>\n",
|
290 |
+
" <tr>\n",
|
291 |
+
" <th>4</th>\n",
|
292 |
+
" <td>longitude</td>\n",
|
293 |
+
" <td>0.362837</td>\n",
|
294 |
+
" </tr>\n",
|
295 |
+
" <tr>\n",
|
296 |
+
" <th>5</th>\n",
|
297 |
+
" <td>living_status_Rent</td>\n",
|
298 |
+
" <td>0.128913</td>\n",
|
299 |
+
" </tr>\n",
|
300 |
+
" <tr>\n",
|
301 |
+
" <th>6</th>\n",
|
302 |
+
" <td>policy_report_filed_ind</td>\n",
|
303 |
+
" <td>0.083922</td>\n",
|
304 |
+
" </tr>\n",
|
305 |
+
" <tr>\n",
|
306 |
+
" <th>7</th>\n",
|
307 |
+
" <td>channel_Phone</td>\n",
|
308 |
+
" <td>0.039526</td>\n",
|
309 |
+
" </tr>\n",
|
310 |
+
" <tr>\n",
|
311 |
+
" <th>8</th>\n",
|
312 |
+
" <td>liab_prct</td>\n",
|
313 |
+
" <td>0.031912</td>\n",
|
314 |
+
" </tr>\n",
|
315 |
+
" <tr>\n",
|
316 |
+
" <th>9</th>\n",
|
317 |
+
" <td>vehicle_weight</td>\n",
|
318 |
+
" <td>0.031770</td>\n",
|
319 |
+
" </tr>\n",
|
320 |
+
" <tr>\n",
|
321 |
+
" <th>10</th>\n",
|
322 |
+
" <td>vehicle_price</td>\n",
|
323 |
+
" <td>0.030162</td>\n",
|
324 |
+
" </tr>\n",
|
325 |
+
" <tr>\n",
|
326 |
+
" <th>11</th>\n",
|
327 |
+
" <td>vehicle_category_Medium</td>\n",
|
328 |
+
" <td>0.027484</td>\n",
|
329 |
+
" </tr>\n",
|
330 |
+
" <tr>\n",
|
331 |
+
" <th>12</th>\n",
|
332 |
+
" <td>vehicle_category_Large</td>\n",
|
333 |
+
" <td>-0.063941</td>\n",
|
334 |
+
" </tr>\n",
|
335 |
+
" <tr>\n",
|
336 |
+
" <th>13</th>\n",
|
337 |
+
" <td>latitude</td>\n",
|
338 |
+
" <td>-0.166059</td>\n",
|
339 |
+
" </tr>\n",
|
340 |
+
" <tr>\n",
|
341 |
+
" <th>14</th>\n",
|
342 |
+
" <td>accident_site_Local</td>\n",
|
343 |
+
" <td>-0.234709</td>\n",
|
344 |
+
" </tr>\n",
|
345 |
+
" <tr>\n",
|
346 |
+
" <th>15</th>\n",
|
347 |
+
" <td>gender_M</td>\n",
|
348 |
+
" <td>-0.277402</td>\n",
|
349 |
+
" </tr>\n",
|
350 |
+
" <tr>\n",
|
351 |
+
" <th>16</th>\n",
|
352 |
+
" <td>channel_Online</td>\n",
|
353 |
+
" <td>-0.306284</td>\n",
|
354 |
+
" </tr>\n",
|
355 |
+
" <tr>\n",
|
356 |
+
" <th>17</th>\n",
|
357 |
+
" <td>claim_est_payout</td>\n",
|
358 |
+
" <td>-0.344002</td>\n",
|
359 |
+
" </tr>\n",
|
360 |
+
" <tr>\n",
|
361 |
+
" <th>18</th>\n",
|
362 |
+
" <td>marital_status</td>\n",
|
363 |
+
" <td>-0.459327</td>\n",
|
364 |
+
" </tr>\n",
|
365 |
+
" <tr>\n",
|
366 |
+
" <th>19</th>\n",
|
367 |
+
" <td>high_education_ind</td>\n",
|
368 |
+
" <td>-0.647302</td>\n",
|
369 |
+
" </tr>\n",
|
370 |
+
" <tr>\n",
|
371 |
+
" <th>20</th>\n",
|
372 |
+
" <td>witness_present_ind</td>\n",
|
373 |
+
" <td>-0.709166</td>\n",
|
374 |
+
" </tr>\n",
|
375 |
+
" <tr>\n",
|
376 |
+
" <th>21</th>\n",
|
377 |
+
" <td>accident_site_Parking Lot</td>\n",
|
378 |
+
" <td>-1.012493</td>\n",
|
379 |
+
" </tr>\n",
|
380 |
+
" <tr>\n",
|
381 |
+
" <th>22</th>\n",
|
382 |
+
" <td>safty_rating</td>\n",
|
383 |
+
" <td>-1.031068</td>\n",
|
384 |
+
" </tr>\n",
|
385 |
+
" <tr>\n",
|
386 |
+
" <th>23</th>\n",
|
387 |
+
" <td>age_of_driver</td>\n",
|
388 |
+
" <td>-2.510087</td>\n",
|
389 |
+
" </tr>\n",
|
390 |
+
" </tbody>\n",
|
391 |
+
"</table>\n",
|
392 |
+
"</div>"
|
393 |
+
],
|
394 |
+
"text/plain": [
|
395 |
+
" feature_name coefficient\n",
|
396 |
+
"0 past_num_of_claims 1.750160\n",
|
397 |
+
"1 annual_income 1.570769\n",
|
398 |
+
"2 age_of_vehicle 0.982407\n",
|
399 |
+
"3 address_change_ind 0.398596\n",
|
400 |
+
"4 longitude 0.362837\n",
|
401 |
+
"5 living_status_Rent 0.128913\n",
|
402 |
+
"6 policy_report_filed_ind 0.083922\n",
|
403 |
+
"7 channel_Phone 0.039526\n",
|
404 |
+
"8 liab_prct 0.031912\n",
|
405 |
+
"9 vehicle_weight 0.031770\n",
|
406 |
+
"10 vehicle_price 0.030162\n",
|
407 |
+
"11 vehicle_category_Medium 0.027484\n",
|
408 |
+
"12 vehicle_category_Large -0.063941\n",
|
409 |
+
"13 latitude -0.166059\n",
|
410 |
+
"14 accident_site_Local -0.234709\n",
|
411 |
+
"15 gender_M -0.277402\n",
|
412 |
+
"16 channel_Online -0.306284\n",
|
413 |
+
"17 claim_est_payout -0.344002\n",
|
414 |
+
"18 marital_status -0.459327\n",
|
415 |
+
"19 high_education_ind -0.647302\n",
|
416 |
+
"20 witness_present_ind -0.709166\n",
|
417 |
+
"21 accident_site_Parking Lot -1.012493\n",
|
418 |
+
"22 safty_rating -1.031068\n",
|
419 |
+
"23 age_of_driver -2.510087"
|
420 |
+
]
|
421 |
+
},
|
422 |
+
"execution_count": 42,
|
423 |
+
"metadata": {},
|
424 |
+
"output_type": "execute_result"
|
425 |
+
}
|
426 |
+
],
|
427 |
+
"source": [
|
428 |
+
"def add_dummies(df, categorical_features):\n",
|
429 |
+
" dummies = pd.get_dummies(df[categorical_features], drop_first=True)\n",
|
430 |
+
" df = pd.concat([dummies, df], axis=1)\n",
|
431 |
+
" df = df.drop(categorical_features, axis=1)\n",
|
432 |
+
" return df.columns\n",
|
433 |
+
"\n",
|
434 |
+
"feature_names = add_dummies(X_train, categorical_features)\n",
|
435 |
+
"\n",
|
436 |
+
"pd.DataFrame({\n",
|
437 |
+
" \"feature_name\": feature_names,\n",
|
438 |
+
" \"coefficient\": lr_pipeline._final_estimator.coef_[0]\n",
|
439 |
+
"}).sort_values(by=\"coefficient\", ascending=False).reset_index(drop=True)"
|
440 |
+
]
|
441 |
+
},
|
442 |
+
{
|
443 |
+
"attachments": {},
|
444 |
+
"cell_type": "markdown",
|
445 |
+
"metadata": {},
|
446 |
+
"source": [
|
447 |
+
"### XGBoost\n",
|
448 |
+
"\n",
|
449 |
+
"Since there are many hyperparameters in XGBoost, I decide to use a randomized search for hyperparameter tuning."
|
450 |
+
]
|
451 |
+
},
|
452 |
+
{
|
453 |
+
"cell_type": "code",
|
454 |
+
"execution_count": 43,
|
455 |
+
"metadata": {},
|
456 |
+
"outputs": [
|
457 |
+
{
|
458 |
+
"name": "stdout",
|
459 |
+
"output_type": "stream",
|
460 |
+
"text": [
|
461 |
+
"Best params: {'subsample': 0.7, 'n_estimators': 100, 'min_child_weight': 7.0, 'max_depth': 1, 'learning_rate': 0.3, 'gamma': 0.25, 'colsample_bytree': 1.0, 'colsample_bylevel': 0.8}\n",
|
462 |
+
"AUC score: 0.7299474921988243\n"
|
463 |
+
]
|
464 |
+
}
|
465 |
+
],
|
466 |
+
"source": [
|
467 |
+
"param_grid = {\n",
|
468 |
+
" \"max_depth\": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],\n",
|
469 |
+
" \"learning_rate\": [0.001, 0.01, 0.1, 0.2, 0.3],\n",
|
470 |
+
" \"subsample\": [0.5, 0.6, 0.7, 0.8, 0.9, 1.0],\n",
|
471 |
+
" \"colsample_bytree\": [0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],\n",
|
472 |
+
" \"colsample_bylevel\": [0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],\n",
|
473 |
+
" \"min_child_weight\": [0.5, 1.0, 3.0, 5.0, 7.0, 10.0],\n",
|
474 |
+
" \"gamma\": [0, 0.25, 0.5, 1.0],\n",
|
475 |
+
" \"n_estimators\": [10, 20, 40, 60, 80, 100, 150, 200]\n",
|
476 |
+
"}\n",
|
477 |
+
"\n",
|
478 |
+
"xgb_clf = RandomizedSearchCV(\n",
|
479 |
+
" XGBClassifier(),\n",
|
480 |
+
" param_distributions=param_grid,\n",
|
481 |
+
" n_iter=50,\n",
|
482 |
+
" n_jobs=-1,\n",
|
483 |
+
" cv=5,\n",
|
484 |
+
" random_state=23,\n",
|
485 |
+
" scoring=\"roc_auc\",\n",
|
486 |
+
")\n",
|
487 |
+
"\n",
|
488 |
+
"xgb_pipeline = modeling(X_train, y_train, X_val, y_val, [column_transformer, scaler, xgb_clf])"
|
489 |
+
]
|
490 |
+
},
|
491 |
+
{
|
492 |
+
"attachments": {},
|
493 |
+
"cell_type": "markdown",
|
494 |
+
"metadata": {},
|
495 |
+
"source": [
|
496 |
+
"Although the class imbalance is not very serious in this dataset, I want to see if using SMOTE to synthesize new examples for the minority class can improve the predictive performance. However, it seems that using SMOTE only worsens the performance."
|
497 |
+
]
|
498 |
+
},
|
499 |
+
{
|
500 |
+
"cell_type": "code",
|
501 |
+
"execution_count": 44,
|
502 |
+
"metadata": {},
|
503 |
+
"outputs": [
|
504 |
+
{
|
505 |
+
"name": "stdout",
|
506 |
+
"output_type": "stream",
|
507 |
+
"text": [
|
508 |
+
"Best params: {'subsample': 1.0, 'n_estimators': 200, 'min_child_weight': 0.5, 'max_depth': 10, 'learning_rate': 0.1, 'gamma': 0.25, 'colsample_bytree': 0.5, 'colsample_bylevel': 0.6}\n",
|
509 |
+
"AUC score: 0.6962796916323821\n"
|
510 |
+
]
|
511 |
+
}
|
512 |
+
],
|
513 |
+
"source": [
|
514 |
+
"sampler = SMOTE(random_state=42)\n",
|
515 |
+
"xgb_pipeline_smote = modeling(X_train, y_train, X_val, y_val, [column_transformer, scaler, sampler, xgb_clf])"
|
516 |
+
]
|
517 |
+
},
|
518 |
+
{
|
519 |
+
"attachments": {},
|
520 |
+
"cell_type": "markdown",
|
521 |
+
"metadata": {},
|
522 |
+
"source": [
|
523 |
+
"Save the XGBoost model (without SMOTE), since it has the best performance."
|
524 |
+
]
|
525 |
+
},
|
526 |
+
{
|
527 |
+
"cell_type": "code",
|
528 |
+
"execution_count": 45,
|
529 |
+
"metadata": {},
|
530 |
+
"outputs": [],
|
531 |
+
"source": [
|
532 |
+
"best_model = xgb_pipeline._final_estimator.best_estimator_\n",
|
533 |
+
"steps = [column_transformer, scaler, best_model]\n",
|
534 |
+
"pipeline = make_pipeline(*steps)\n",
|
535 |
+
"y_test_pred = pipeline.predict_proba(X_test)[:, 1]\n",
|
536 |
+
"\n",
|
537 |
+
"df = pd.DataFrame({\n",
|
538 |
+
" \"claim_number\": df_test[\"claim_number\"],\n",
|
539 |
+
" \"fraud\": y_test_pred\n",
|
540 |
+
"})\n",
|
541 |
+
"#df.to_csv(\"../data/submission/submission.csv\", index=False)"
|
542 |
+
]
|
543 |
+
},
|
544 |
+
{
|
545 |
+
"attachments": {},
|
546 |
+
"cell_type": "markdown",
|
547 |
+
"metadata": {},
|
548 |
+
"source": [
|
549 |
+
"To examine which feature is important, I introduce a feature with random numbers. A feature can be considered as important If the importance of that feature is larger than that of the random feature."
|
550 |
+
]
|
551 |
+
},
|
552 |
+
{
|
553 |
+
"cell_type": "code",
|
554 |
+
"execution_count": 18,
|
555 |
+
"metadata": {},
|
556 |
+
"outputs": [
|
557 |
+
{
|
558 |
+
"data": {
|
559 |
+
"text/html": [
|
560 |
+
"<div>\n",
|
561 |
+
"<style scoped>\n",
|
562 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
563 |
+
" vertical-align: middle;\n",
|
564 |
+
" }\n",
|
565 |
+
"\n",
|
566 |
+
" .dataframe tbody tr th {\n",
|
567 |
+
" vertical-align: top;\n",
|
568 |
+
" }\n",
|
569 |
+
"\n",
|
570 |
+
" .dataframe thead th {\n",
|
571 |
+
" text-align: right;\n",
|
572 |
+
" }\n",
|
573 |
+
"</style>\n",
|
574 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
575 |
+
" <thead>\n",
|
576 |
+
" <tr style=\"text-align: right;\">\n",
|
577 |
+
" <th></th>\n",
|
578 |
+
" <th>feature_name</th>\n",
|
579 |
+
" <th>importance</th>\n",
|
580 |
+
" </tr>\n",
|
581 |
+
" </thead>\n",
|
582 |
+
" <tbody>\n",
|
583 |
+
" <tr>\n",
|
584 |
+
" <th>0</th>\n",
|
585 |
+
" <td>accident_site_Parking Lot</td>\n",
|
586 |
+
" <td>0.111572</td>\n",
|
587 |
+
" </tr>\n",
|
588 |
+
" <tr>\n",
|
589 |
+
" <th>1</th>\n",
|
590 |
+
" <td>high_education_ind</td>\n",
|
591 |
+
" <td>0.082720</td>\n",
|
592 |
+
" </tr>\n",
|
593 |
+
" <tr>\n",
|
594 |
+
" <th>2</th>\n",
|
595 |
+
" <td>witness_present_ind</td>\n",
|
596 |
+
" <td>0.072724</td>\n",
|
597 |
+
" </tr>\n",
|
598 |
+
" <tr>\n",
|
599 |
+
" <th>3</th>\n",
|
600 |
+
" <td>past_num_of_claims</td>\n",
|
601 |
+
" <td>0.052461</td>\n",
|
602 |
+
" </tr>\n",
|
603 |
+
" <tr>\n",
|
604 |
+
" <th>4</th>\n",
|
605 |
+
" <td>marital_status</td>\n",
|
606 |
+
" <td>0.052270</td>\n",
|
607 |
+
" </tr>\n",
|
608 |
+
" <tr>\n",
|
609 |
+
" <th>5</th>\n",
|
610 |
+
" <td>address_change_ind</td>\n",
|
611 |
+
" <td>0.044381</td>\n",
|
612 |
+
" </tr>\n",
|
613 |
+
" <tr>\n",
|
614 |
+
" <th>6</th>\n",
|
615 |
+
" <td>age_of_driver</td>\n",
|
616 |
+
" <td>0.039922</td>\n",
|
617 |
+
" </tr>\n",
|
618 |
+
" <tr>\n",
|
619 |
+
" <th>7</th>\n",
|
620 |
+
" <td>longitude</td>\n",
|
621 |
+
" <td>0.034581</td>\n",
|
622 |
+
" </tr>\n",
|
623 |
+
" <tr>\n",
|
624 |
+
" <th>8</th>\n",
|
625 |
+
" <td>safty_rating</td>\n",
|
626 |
+
" <td>0.033645</td>\n",
|
627 |
+
" </tr>\n",
|
628 |
+
" <tr>\n",
|
629 |
+
" <th>9</th>\n",
|
630 |
+
" <td>claim_est_payout</td>\n",
|
631 |
+
" <td>0.032631</td>\n",
|
632 |
+
" </tr>\n",
|
633 |
+
" <tr>\n",
|
634 |
+
" <th>10</th>\n",
|
635 |
+
" <td>random_feature</td>\n",
|
636 |
+
" <td>0.032600</td>\n",
|
637 |
+
" </tr>\n",
|
638 |
+
" <tr>\n",
|
639 |
+
" <th>11</th>\n",
|
640 |
+
" <td>liab_prct</td>\n",
|
641 |
+
" <td>0.032246</td>\n",
|
642 |
+
" </tr>\n",
|
643 |
+
" <tr>\n",
|
644 |
+
" <th>12</th>\n",
|
645 |
+
" <td>vehicle_price</td>\n",
|
646 |
+
" <td>0.032152</td>\n",
|
647 |
+
" </tr>\n",
|
648 |
+
" <tr>\n",
|
649 |
+
" <th>13</th>\n",
|
650 |
+
" <td>annual_income</td>\n",
|
651 |
+
" <td>0.031335</td>\n",
|
652 |
+
" </tr>\n",
|
653 |
+
" <tr>\n",
|
654 |
+
" <th>14</th>\n",
|
655 |
+
" <td>vehicle_weight</td>\n",
|
656 |
+
" <td>0.030896</td>\n",
|
657 |
+
" </tr>\n",
|
658 |
+
" <tr>\n",
|
659 |
+
" <th>15</th>\n",
|
660 |
+
" <td>latitude</td>\n",
|
661 |
+
" <td>0.030324</td>\n",
|
662 |
+
" </tr>\n",
|
663 |
+
" <tr>\n",
|
664 |
+
" <th>16</th>\n",
|
665 |
+
" <td>channel_Online</td>\n",
|
666 |
+
" <td>0.030144</td>\n",
|
667 |
+
" </tr>\n",
|
668 |
+
" <tr>\n",
|
669 |
+
" <th>17</th>\n",
|
670 |
+
" <td>accident_site_Local</td>\n",
|
671 |
+
" <td>0.029325</td>\n",
|
672 |
+
" </tr>\n",
|
673 |
+
" <tr>\n",
|
674 |
+
" <th>18</th>\n",
|
675 |
+
" <td>gender_M</td>\n",
|
676 |
+
" <td>0.028732</td>\n",
|
677 |
+
" </tr>\n",
|
678 |
+
" <tr>\n",
|
679 |
+
" <th>19</th>\n",
|
680 |
+
" <td>vehicle_category_Large</td>\n",
|
681 |
+
" <td>0.028661</td>\n",
|
682 |
+
" </tr>\n",
|
683 |
+
" <tr>\n",
|
684 |
+
" <th>20</th>\n",
|
685 |
+
" <td>channel_Phone</td>\n",
|
686 |
+
" <td>0.027671</td>\n",
|
687 |
+
" </tr>\n",
|
688 |
+
" <tr>\n",
|
689 |
+
" <th>21</th>\n",
|
690 |
+
" <td>vehicle_category_Medium</td>\n",
|
691 |
+
" <td>0.027547</td>\n",
|
692 |
+
" </tr>\n",
|
693 |
+
" <tr>\n",
|
694 |
+
" <th>22</th>\n",
|
695 |
+
" <td>living_status_Rent</td>\n",
|
696 |
+
" <td>0.027294</td>\n",
|
697 |
+
" </tr>\n",
|
698 |
+
" <tr>\n",
|
699 |
+
" <th>23</th>\n",
|
700 |
+
" <td>age_of_vehicle</td>\n",
|
701 |
+
" <td>0.027125</td>\n",
|
702 |
+
" </tr>\n",
|
703 |
+
" <tr>\n",
|
704 |
+
" <th>24</th>\n",
|
705 |
+
" <td>policy_report_filed_ind</td>\n",
|
706 |
+
" <td>0.027040</td>\n",
|
707 |
+
" </tr>\n",
|
708 |
+
" </tbody>\n",
|
709 |
+
"</table>\n",
|
710 |
+
"</div>"
|
711 |
+
],
|
712 |
+
"text/plain": [
|
713 |
+
" feature_name importance\n",
|
714 |
+
"0 accident_site_Parking Lot 0.111572\n",
|
715 |
+
"1 high_education_ind 0.082720\n",
|
716 |
+
"2 witness_present_ind 0.072724\n",
|
717 |
+
"3 past_num_of_claims 0.052461\n",
|
718 |
+
"4 marital_status 0.052270\n",
|
719 |
+
"5 address_change_ind 0.044381\n",
|
720 |
+
"6 age_of_driver 0.039922\n",
|
721 |
+
"7 longitude 0.034581\n",
|
722 |
+
"8 safty_rating 0.033645\n",
|
723 |
+
"9 claim_est_payout 0.032631\n",
|
724 |
+
"10 random_feature 0.032600\n",
|
725 |
+
"11 liab_prct 0.032246\n",
|
726 |
+
"12 vehicle_price 0.032152\n",
|
727 |
+
"13 annual_income 0.031335\n",
|
728 |
+
"14 vehicle_weight 0.030896\n",
|
729 |
+
"15 latitude 0.030324\n",
|
730 |
+
"16 channel_Online 0.030144\n",
|
731 |
+
"17 accident_site_Local 0.029325\n",
|
732 |
+
"18 gender_M 0.028732\n",
|
733 |
+
"19 vehicle_category_Large 0.028661\n",
|
734 |
+
"20 channel_Phone 0.027671\n",
|
735 |
+
"21 vehicle_category_Medium 0.027547\n",
|
736 |
+
"22 living_status_Rent 0.027294\n",
|
737 |
+
"23 age_of_vehicle 0.027125\n",
|
738 |
+
"24 policy_report_filed_ind 0.027040"
|
739 |
+
]
|
740 |
+
},
|
741 |
+
"execution_count": 18,
|
742 |
+
"metadata": {},
|
743 |
+
"output_type": "execute_result"
|
744 |
+
}
|
745 |
+
],
|
746 |
+
"source": [
|
747 |
+
"X_train[\"random_feature\"] = np.random.uniform(size=len(X_train))\n",
|
748 |
+
"xgb_clf_random_feature = XGBClassifier(**xgb_pipeline._final_estimator.best_params_)\n",
|
749 |
+
"steps = [column_transformer, scaler, xgb_clf_random_feature]\n",
|
750 |
+
"xgb_pipeline_random_feature = make_pipeline(*steps)\n",
|
751 |
+
"xgb_pipeline_random_feature = xgb_pipeline_random_feature.fit(X_train, y_train)\n",
|
752 |
+
"\n",
|
753 |
+
"pd.DataFrame({\n",
|
754 |
+
" \"feature_name\": list(feature_names) + [\"random_feature\"],\n",
|
755 |
+
" \"importance\": xgb_pipeline_random_feature._final_estimator.feature_importances_\n",
|
756 |
+
"}).sort_values(by=\"importance\", ascending=False).reset_index(drop=True)"
|
757 |
+
]
|
758 |
+
},
|
759 |
+
{
|
760 |
+
"cell_type": "code",
|
761 |
+
"execution_count": null,
|
762 |
+
"metadata": {},
|
763 |
+
"outputs": [],
|
764 |
+
"source": [
|
765 |
+
"X_train"
|
766 |
+
]
|
767 |
+
},
|
768 |
+
{
|
769 |
+
"cell_type": "code",
|
770 |
+
"execution_count": null,
|
771 |
+
"metadata": {},
|
772 |
+
"outputs": [],
|
773 |
+
"source": [
|
774 |
+
"y_train"
|
775 |
+
]
|
776 |
+
},
|
777 |
+
{
|
778 |
+
"cell_type": "code",
|
779 |
+
"execution_count": 47,
|
780 |
+
"metadata": {},
|
781 |
+
"outputs": [],
|
782 |
+
"source": [
|
783 |
+
"with open('./best_model_3.pickle', 'wb') as handle:\n",
|
784 |
+
" #pickle.dump(a, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
|
785 |
+
"\n",
|
786 |
+
" pickle.dump(xgb_pipeline, handle)"
|
787 |
+
]
|
788 |
+
},
|
789 |
+
{
|
790 |
+
"cell_type": "code",
|
791 |
+
"execution_count": 31,
|
792 |
+
"metadata": {},
|
793 |
+
"outputs": [],
|
794 |
+
"source": [
|
795 |
+
"import pickle"
|
796 |
+
]
|
797 |
+
},
|
798 |
+
{
|
799 |
+
"cell_type": "code",
|
800 |
+
"execution_count": null,
|
801 |
+
"metadata": {},
|
802 |
+
"outputs": [],
|
803 |
+
"source": []
|
804 |
+
}
|
805 |
+
],
|
806 |
+
"metadata": {
|
807 |
+
"interpreter": {
|
808 |
+
"hash": "03e93f2959c516196957ae17ec0aa5d1e9fc5dd82cbe13968d4cfc2a60558992"
|
809 |
+
},
|
810 |
+
"kernelspec": {
|
811 |
+
"display_name": "Python 3 (ipykernel)",
|
812 |
+
"language": "python",
|
813 |
+
"name": "python3"
|
814 |
+
},
|
815 |
+
"language_info": {
|
816 |
+
"codemirror_mode": {
|
817 |
+
"name": "ipython",
|
818 |
+
"version": 3
|
819 |
+
},
|
820 |
+
"file_extension": ".py",
|
821 |
+
"mimetype": "text/x-python",
|
822 |
+
"name": "python",
|
823 |
+
"nbconvert_exporter": "python",
|
824 |
+
"pygments_lexer": "ipython3",
|
825 |
+
"version": "3.12.1"
|
826 |
+
}
|
827 |
+
},
|
828 |
+
"nbformat": 4,
|
829 |
+
"nbformat_minor": 4
|
830 |
+
}
|