Skip to main content

​
Overview

Feature engineering transforms raw data into meaningful features that improve model performance. This process includes date extraction, categorical encoding, and feature scaling.

​
Date Feature Extraction

​
Converting to DateTime

First, date columns are converted from strings to datetime objects: 
full_dataset_preprocessed['Created Date'] = pd.to_datetime(
    full_dataset_preprocessed['Created Date'], 
    format="%Y-%m-%d"
)
full_dataset_preprocessed['Close Date'] = pd.to_datetime(
    full_dataset_preprocessed['Close Date'], 
    format="%Y-%m-%d"
)

​
Extracting Year and Month

Temporal features are extracted from the datetime columns: 
full_dataset_preprocessed['Created Year'] = full_dataset_preprocessed['Created Date'].dt.year
full_dataset_preprocessed['Created Month'] = full_dataset_preprocessed['Created Date'].dt.month
full_dataset_preprocessed['Close Year'] = full_dataset_preprocessed['Close Date'].dt.year
full_dataset_preprocessed['Close Month'] = full_dataset_preprocessed['Close Date'].dt.month
This creates four new features:
  • Created Year: Year when the offer was created
  • Created Month: Month when the offer was created
  • Close Year: Year when the offer was closed
  • Close Month: Month when the offer was closed
Extracting year and month as separate features allows the model to capture seasonal patterns and temporal trends in lead conversion.

​
Dropping Original Date Columns

After extraction, the original datetime columns are removed: 
full_dataset_preprocessed = full_dataset_preprocessed.drop(
    ['Created Date', 'Close Date'], 
    axis=1
)

​
Target Variable Mapping

​
Status Column Transformation

The Status column contains the target variable. Minority classes are grouped to address class imbalance: 
clase_mapping = {'Closed Won': 'Closed Won', 'Closed Lost': 'Closed Lost'}
# Assign 'Other' to all classes that are not 'Closed Won' or 'Closed Lost'
full_dataset_preprocessed['Status'] = full_dataset_preprocessed['Status'].map(clase_mapping).fillna('Other')
This creates three target classes:
  • Closed Won: Successfully converted leads
  • Closed Lost: Lost opportunities
  • Other: All other statuses (e.g., In Progress, Nurturing)
Grouping minority classes into “Other” helps prevent overfitting on rare categories and improves model generalization.

​
Label Encoding

​
Why Label Encoding?

Machine learning models require numerical inputs. Label Encoding converts categorical variables into integer representations.

​
Implementation

1

Create LabelEncoder instance

from sklearn.preprocessing import LabelEncoder

# Creating a instance of label Encoder.
label_encoder = LabelEncoder()
2

Identify categorical columns

Select all columns with object or datetime data types:
# List of categorical columns to encode
categorical_columns = full_dataset_preprocessed.select_dtypes(
    ['object', 'datetime64[ns]']
).columns
categorical_columns = list(set(categorical_columns))
3

Apply encoding

Transform each categorical column into numerical values:
# Aplicar LabelEncoder a cada columna categórica
for column in categorical_columns:
    if column in full_dataset_preprocessed.columns:
        full_dataset_preprocessed[column] = label_encoder.fit_transform(
            full_dataset_preprocessed[column]
        )

​
Encoded Features

The following categorical features are encoded:
  • Source (Inbound, Outbound, etc.)
  • City (Various cities)
  • Loss Reason (Reasons for lost opportunities)
  • Pain (Customer pain level)
  • Discount code (Applied discount codes)
  • Status (Closed Won, Closed Lost, Other)
  • Use Case (Type of use case)
Each unique category is assigned a unique integer. For example, if Source has values [“Inbound”, “Outbound”, “Referral”], they might be encoded as [0, 1, 2].

​
Data Scaling

After encoding, numerical features are scaled using StandardScaler to normalize their ranges: 
from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
# Scale numerical features
StandardScaler transforms features to have zero mean and unit variance, which helps gradient boosting algorithms converge faster and perform better.

​
Final Output

The fully preprocessed dataset is saved for model training: 
full_dataset_preprocessed.to_csv("data/processed/full_dataset.csv", index=False)

​
Summary

The feature engineering pipeline:
  1. Extracts temporal features (year, month) from dates
  2. Maps target variable to three balanced classes
  3. Encodes all categorical variables using LabelEncoder
  4. Scales numerical features using StandardScaler
  5. Outputs a clean, model-ready dataset
This processed dataset is now ready for training the Gradient Boosting classifier.

Build docs developers (and LLMs) love

Get started for freeTalk to us