src/data/preprocessing.py - master - LingLing/loan-default-prediction

src/data/preprocessing.py
56 строк2 KB
liza
Project update
06 авг 2026, 20:20
06 авг 2026, 20:20b55a4f3
import pandas as pd
import numpy as np
from sklearn.base import BaseEstimator, TransformerMixin


class Preprocessor(BaseEstimator, TransformerMixin):

    def __init__(self):

        self.extreme_utilization_threshold = 5.0
        self.extreme_debt_threshold = 10.0
        self.extreme_realastate_threshold = 5.0

        self.age_median = None

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

        self.NumberOfTime30_59DaysPastDueNotWorse_ = X['NumberOfTime30-59DaysPastDueNotWorse'].quantile(0.99)
        self.NumberOfTime60_89DaysPastDueNotWorse_ = X['NumberOfTime60-89DaysPastDueNotWorse'].quantile(0.99)
        self.NumberOfTimes90DaysLate_ = X['NumberOfTimes90DaysLate'].quantile(0.99)

        self.age_median_ = X['age'].median()

        return self

    def transform(self, X, y=None):

        X = X.copy()
        base_X = X.copy()

        X.loc[X['age'] < 21, 'age'] = self.age_median_

        columns = ['NumberOfTime30-59DaysPastDueNotWorse',
                   'NumberOfTimes90DaysLate',
                   'NumberOfTime60-89DaysPastDueNotWorse']

        anom_mask = base_X[columns].apply(lambda row: row.nunique() == 1, axis=1) & base_X[columns].isin([96, 98]).any(axis=1)

        X.loc[anom_mask, 'NumberOfTime30-59DaysPastDueNotWorse'] = self.NumberOfTime30_59DaysPastDueNotWorse_
        X.loc[anom_mask, 'NumberOfTime60-89DaysPastDueNotWorse'] = self.NumberOfTime60_89DaysPastDueNotWorse_
        X.loc[anom_mask, 'NumberOfTimes90DaysLate'] = self.NumberOfTimes90DaysLate_

        X['RevolvingUtilization_clean'] = base_X['RevolvingUtilizationOfUnsecuredLines'].clip(upper=self.extreme_utilization_threshold)

        X['DebtRatio_capped'] = base_X['DebtRatio'].clip(upper=self.extreme_debt_threshold)

        X['Realestate_capped'] = base_X['NumberRealEstateLoansOrLines'].clip(upper=self.extreme_realastate_threshold)

        cols_to_drop = [
            'RevolvingUtilizationOfUnsecuredLines',
            'DebtRatio',
            'NumberRealEstateLoansOrLines'
        ]
        X = X.drop(columns=cols_to_drop, errors='ignore')

        return X