Computational Thinking and Logic Building Classes in Binh Duong

1. Why Clean Data Matters

Garbage In, Garbage Out (GIGO): models trained on noisy or incomplete data perform poorly. This project prepares a 1,000-row loan dataset with EDA, imputation, outlier clipping, and domain features for credit risk modeling.

2. Load and Impute

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

df = pd.read_csv("loan_risk_data.csv")
print(df.shape)  # (1000, 6)
print(df.isnull().sum())  # Income: 51 missing

income_median = df["Income"].median()
df["Income"] = df["Income"].fillna(income_median)

Columns: Loan_ID, Age, Income, Loan_Amount, Credit_Score, Default. Median fills the 51 missing Income values so all six columns are complete.

3. Summaries and Correlation

print(df.describe())  # Loan_Amount Q3 ~21,000 vs max 500,000
sns.heatmap(df.corr(), annot=True, cmap="coolwarm", fmt=".2f")
plt.show()

Large Q3–max gap signals outliers. Loan_Amount vs Loan_ID correlation is about 0.38; other pairs are weak.

4. Clip Outliers

sns.boxplot(x=df["Loan_Amount"])  # extreme point at 500,000
upper_limit = df["Loan_Amount"].quantile(0.90)
df["Loan_Amount"] = df["Loan_Amount"].clip(upper=upper_limit)
sns.boxplot(x=df["Loan_Amount"])  # cleaned distribution

Clipping caps extreme amounts without dropping applicant rows.

5. Feature Engineering

df["DTI_Ratio"] = df["Loan_Amount"] / df["Income"]

bins = [0, 580, 670, 740, 850]  # example boundaries
labels = ["High Risk", "Fair", "Good", "Excellent"]
df["Risk_Category"] = pd.cut(df["Credit_Score"], bins=bins, labels=labels)

print(df[["Income", "DTI_Ratio", "Risk_Category"]].head())

• DTI_Ratio = Loan_Amount / Income — repayment burden
• Risk_Category — credit score binned into High Risk, Fair, Good, Excellent

6. Portfolio Snapshot

• High Risk ~19.3%
• Fair ~20.8%
• Good ~19.25%
• Excellent ~23.0%

After imputation, clipping, and engineered features, the dataset is ready for model training.