full

from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.ensemble import RandomForestClassifier
from sklearn.preprocessing import LabelEncoder
import pandas as pd
import numpy as np

class FeatureSelector(BaseEstimator, TransformerMixin):
    def __init__(self, feature_name):
        self.feature_name = feature_name

    def fit(self, X, y=None):
        return self

    def transform(self, X):
        return X.loc[:, self.feature_name]

class CustomBinTransformer(BaseEstimator, TransformerMixin):
    def __init__(self, ranges):
        self.ranges = ranges

    def fit(self, X, y=None):
        return self

    def transform(self, X):
        # тут пишешь деление по бинам, можешь использовать стандартные, если подойдут из sklearn. Пишу лишь как пример:
        return pd.cut(X, self.ranges, labels=False).values[:, np.newaxis]


model = RandomForestClassifier() # Тут для примера твоя модель

features = FeatureUnion([
    ('НБКИ+', Pipeline([
            ('selector', FeatureSelector('НБКИ+')),
            ('bining', CustomBinTransformer([0, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000])) # Списком просто передаю диапазноы для бининга
        ])
    ),
    ('ОКБ+', Pipeline([
            ('selector', FeatureSelector('ОКБ+')),
            ('bining', CustomBinTransformer([0, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000]))
        ])
    ),
    ('Эквифакс+', Pipeline([
            ('selector', FeatureSelector('Эквифакс+')),
            ('bining', CustomBinTransformer([0, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000]))
        ])
    ),
    ('mail.ru+', Pipeline([
            ('selector', FeatureSelector('mail.ru+')),
            ('bining', CustomBinTransformer([0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1]))
        ])
    ),
])

pipeline =  Pipeline([
    ('features', features),
    ('estimator', model)
])
df = pd.DataFrame(np.random.rand(10, 5), columns=['НБКИ+', 'ОКБ+', 'Эквифакс+', 'mail.ru+', 'target'])
df.iloc[:, :3] = df.iloc[:, :3] * 1000
df['target'] = (df['target'] < 0.5).astype(np.int)
pipeline.fit(df, df['target'])
pipeline.predict(df)