Data Preparation for Feature Selection

Welcome to the first lesson in our course on Feature Selection, Reduction and Streamlining. In this unit, we're setting the stage for your journey into feature selection by ensuring your dataset is primed and ready. We'll revisit familiar data preparation techniques, this time with a sharper focus on feature selection. Using the Titanic dataset as our example, remember that well-prepared data is the linchpin to effective feature selection and successful analysis.

You’re already skilled at managing missing data, and it's time to apply this to the Titanic dataset. Ensuring data integrity begins with filling in gaps using statistical measures, such as the median for the age column and the mode for categorical columns like embarked and embark_town. Refresh your technique with these steps:

import pandas as pd

# Load the dataset
df = pd.read_csv("titanic.csv")

# Handle missing values
df['age'] = df['age'].fillna(df['age'].median())
df['embarked'] = df['embarked'].fillna(df['embarked'].mode()[0])
df['embark_town'] = df['embark_town'].fillna(df['embark_town'].mode()[0])

When a column in your dataset has an extensive amount of missing values, it can introduce more challenges than benefits. The deck column in the Titanic dataset is such an example. In cases where the missing data is too significant to handle through imputation, it's often more practical to remove the column entirely. This decision helps maintain the dataset's integrity for analysis.

You can accomplish this using the drop method in Pandas. Here is how you can do it:

# Drop 'deck' column due to excessive missing values
df.drop(columns=['deck'], inplace=True)

The drop method allows us to specify which columns to remove. By setting the inplace=True parameter, the method updates the original DataFrame, meaning the deck column is removed permanently without needing to create a new DataFrame. This approach keeps your code concise and the data management straightforward.

For the Titanic dataset, we're choosing Label Encoding to handle categorical data. This method is suitable when you want to keep the dataset compact without adding too many new columns, as can happen with One-Hot Encoding. Label Encoding transforms each category into a unique integer, simplifying the data while preserving its meaning.

To apply Label Encoding, we use a for loop to go through each categorical column and replace the text categories with numbers:

from sklearn.preprocessing import LabelEncoder

# Initialize the label encoder
label_encoder = LabelEncoder()

# Identify categorical features and apply label encoding
label_columns = df.select_dtypes(include=['object']).columns
for column in label_columns:
    df[column] = label_encoder.fit_transform(df[column])

By working through each column with a loop, we efficiently replace the categories with numerical values, making the dataset ready for machine learning models without increasing its size unnecessarily.

Now that you’ve polished your dataset, it’s crucial to save your work, a step that's new for us. By storing the processed data, you eliminate the need to repeat preparations each time you access the dataset. Here’s the simple way to do it with Pandas:

# Save the updated dataset
df.to_csv("titanic_updated.csv", index=False)

By executing this command, your hard work is preserved in a CSV file, named "titanic_updated.csv" ready for ongoing feature selection efforts.

After saving your polished dataset, it's important to verify the saved file to ensure that all the changes have been successfully recorded. You can open the saved CSV file and display the first few rows to confirm:

# Read the updated dataset
df_updated = pd.read_csv("titanic_updated.csv")

# Display the first few rows
print(df_updated.head())

This should produce the following output, showing the first few entries of your updated dataset:

   survived  pclass  sex   age  ...  adult_male  embark_town  alive  alone
0         0       3    1  22.0  ...        True            2      0  False
1         1       1    0  38.0  ...       False            0      1  False
2         1       3    0  26.0  ...       False            2      1   True
3         1       1    0  35.0  ...       False            2      1  False
4         0       3    1  35.0  ...        True            2      0   True

By running this code, you'll load the saved dataset and display the first few entries, giving you a quick visual confirmation that your data preparation steps were correctly applied and saved.

You've seamlessly reviewed essential steps in preparing your dataset, setting a solid groundwork for effective feature selection. Your adeptness at handling missing values, intelligently dropping columns, encoding categorical features, and saving your datasets forms the foundation for what's to come. As we advance, we'll delve into selecting and streamlining the most impactful features. Upcoming exercises will reinforce these techniques, propelling you toward mastering feature selection.