tldr

KggleのAfrica Economic, Banking and Systemic Crisis DataをBanking Crisis Prediction - Data Every Day #043に沿ってやっていきます。

実行環境はGoogle Colaboratorです。

インポート

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

import sklearn.preprocessing as sp
from sklearn.model_selection import train_test_split
import sklearn.linear_model as slm

import tensorflow as tf

データのダウンロード

Google Driveをマウントします。

from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive

KaggleのAPIクライアントを初期化し、認証します。
認証情報はGoogle Drive内（/content/drive/My Drive/Colab Notebooks/Kaggle）にkaggle.jsonとして置いてあります。

import os
kaggle_path = "/content/drive/My Drive/Colab Notebooks/Kaggle"
os.environ['KAGGLE_CONFIG_DIR'] = kaggle_path

from kaggle.api.kaggle_api_extended import KaggleApi
api = KaggleApi()
api.authenticate()

Kaggle APIを使ってデータをダウンロードします。

dataset_id = 'chirin/africa-economic-banking-and-systemic-crisis-data'
dataset = api.dataset_list_files(dataset_id)
file_name = dataset.files[0].name
file_path = os.path.join(api.get_default_download_dir(), file_name)
file_path

'/content/african_crises.csv'

api.dataset_download_file(dataset_id, file_name, force=True, quiet=False)

100%|██████████| 65.4k/65.4k [00:00<00:00, 27.5MB/s]

Downloading african_crises.csv to /content









True

データの読み込み

Pedumagalhaes/quality-prediction-in-a-mining-processadasを使ってダウンロードしてきたCSVファイルを読み込みます。

data = pd.read_csv(file_path)

data

	case	cc3	country	year	systemic_crisis	exch_usd	domestic_debt_in_default	sovereign_external_debt_default	gdp_weighted_default	inflation_annual_cpi	independence	currency_crises	inflation_crises	banking_crisis
0	1	DZA	Algeria	1870	1	0.052264	0	0	0.0	3.441456	0	0	0	crisis
1	1	DZA	Algeria	1871	0	0.052798	0	0	0.0	14.149140	0	0	0	no_crisis
2	1	DZA	Algeria	1872	0	0.052274	0	0	0.0	-3.718593	0	0	0	no_crisis
3	1	DZA	Algeria	1873	0	0.051680	0	0	0.0	11.203897	0	0	0	no_crisis
4	1	DZA	Algeria	1874	0	0.051308	0	0	0.0	-3.848561	0	0	0	no_crisis
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1054	70	ZWE	Zimbabwe	2009	1	354.800000	1	1	0.0	-7.670000	1	1	0	crisis
1055	70	ZWE	Zimbabwe	2010	0	378.200000	1	1	0.0	3.217000	1	0	0	no_crisis
1056	70	ZWE	Zimbabwe	2011	0	361.900000	1	1	0.0	4.920000	1	0	0	no_crisis
1057	70	ZWE	Zimbabwe	2012	0	361.900000	1	1	0.0	3.720000	1	0	0	no_crisis
1058	70	ZWE	Zimbabwe	2013	0	361.900000	1	1	0.0	1.632000	1	0	0	no_crisis

1059 rows × 14 columns

下準備

data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1059 entries, 0 to 1058
Data columns (total 14 columns):
 #   Column                           Non-Null Count  Dtype  
---  ------                           --------------  -----  
 0   case                             1059 non-null   int64  
 1   cc3                              1059 non-null   object 
 2   country                          1059 non-null   object 
 3   year                             1059 non-null   int64  
 4   systemic_crisis                  1059 non-null   int64  
 5   exch_usd                         1059 non-null   float64
 6   domestic_debt_in_default         1059 non-null   int64  
 7   sovereign_external_debt_default  1059 non-null   int64  
 8   gdp_weighted_default             1059 non-null   float64
 9   inflation_annual_cpi             1059 non-null   float64
 10  independence                     1059 non-null   int64  
 11  currency_crises                  1059 non-null   int64  
 12  inflation_crises                 1059 non-null   int64  
 13  banking_crisis                   1059 non-null   object 
dtypes: float64(3), int64(8), object(3)
memory usage: 116.0+ KB

data.isna().sum()

case                               0
cc3                                0
country                            0
year                               0
systemic_crisis                    0
exch_usd                           0
domestic_debt_in_default           0
sovereign_external_debt_default    0
gdp_weighted_default               0
inflation_annual_cpi               0
independence                       0
currency_crises                    0
inflation_crises                   0
banking_crisis                     0
dtype: int64

エンコード

data = data.drop(['case', 'country'], axis=1)

def onehot_ende(df, column):
    df = df.copy()
    dummies = pd.get_dummies(df[column])
    df = pd.concat([df, dummies], axis=1)
    df = df.drop(column, axis=1)
    return df

data = onehot_ende(data, 'cc3')

y = data['banking_crisis']
X = data.drop(['banking_crisis'], axis=1)

label_encoder = sp.LabelEncoder()
y = label_encoder.fit_transform(y)

{index: label for index, label in enumerate(label_encoder.classes_)}

{0: 'crisis', 1: 'no_crisis'}

pd.Series(y).apply(lambda x: 1 - x)

0       1
1       0
2       0
3       0
4       0
       ..
1054    1
1055    0
1056    0
1057    0
1058    0
Length: 1059, dtype: int64

スケーリング

scaler = sp.StandardScaler()
X = pd.DataFrame(scaler.fit_transform(X), index=X.index, columns=X.columns)

分割

X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7)

Training

X_train.shape

(741, 23)

model = tf.keras.Sequential([
    tf.keras.layers.Dense(64, activation='relu', input_shape=(23,)),
    tf.keras.layers.Dense(64, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid'),
])

model.summary()

model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=[tf.keras.metrics.AUC(name='auc')],
)

batch_size = 64
epochs = 60

history = model.fit(
    X_train,
    y_train,
    validation_split=0.2,
    batch_size=batch_size,
    epochs=epochs,
    callbacks=[tf.keras.callbacks.ReduceLROnPlateau()],
    verbose=0,
)

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_6 (Dense)              (None, 64)                1536      
_________________________________________________________________
dense_7 (Dense)              (None, 64)                4160      
_________________________________________________________________
dense_8 (Dense)              (None, 1)                 65        
=================================================================
Total params: 5,761
Trainable params: 5,761
Non-trainable params: 0
_________________________________________________________________

結果

plt.figure(figsize=(14, 10))

epochs_range = range(1, epochs+1)
train_loss = history.history['loss']
val_loss = history.history['val_loss']

plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')

plt.title('Training and Validation Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()

plt.show()

model.evaluate(X_test, y_test)

10/10 [==============================] - 0s 3ms/step - loss: 0.0758 - auc: 0.9855





[0.07584450393915176, 0.9855029582977295]