4. Before the Model: Essential Steps of Data Preprocessing in ML

When working on machine learning projects, we often get excited about building and training models. But here’s the truth: your model is only as good as your data.

If the data is noisy, inconsistent, or incomplete, no algorithm—no matter how powerful—will give you reliable results. This is where data preprocessing comes in.

In this article, we’ll walk through the essential steps of preparing your data before model training.

---

Why Data Preprocessing Matters

The quality of your dataset directly affects:

• Model performance (better accuracy and predictions)

• Training efficiency (less complexity for the algorithm)

• Interpretability (cleaner results that make sense)

That’s why preprocessing is often said to take 80% of a data scientist’s time.

Commonly used libraries for preprocessing:

Pandas – for data cleaning and manipulation

NumPy – for numerical computations

Key Steps in Data Preprocessing

1. Handling Missing Data

Real-world data is often incomplete. For example, in a dataset of student records, some students might have missing grades. We can’t just feed missing values into a model.

Ways to handle missing data:

• Replace missing data with mean, median, or mode

• Use advanced imputation methods

• Drop rows or columns if too many values are missing

# Example: Fill missing values with the column mean
df['Age'].fillna(df['Age'].mean(), inplace=True)

# Or drop rows with missing values
df.dropna(inplace=True)

2. Detecting and Treating Outliers

Outliers are unusual values that don’t fit the general pattern. For example:

• Age = 500 years in a human dataset.

Why remove them?

• They can skew results and make models less accurate.

Techniques include:

• Using boxplots or z-scores to detect outliers

• Removing or adjusting them

3. Data Cleaning

Data may have duplicates, typos, or irrelevant values. For instance, two identical rows in a sales dataset can bias results.

Steps:

• Remove duplicates

• Standardize formats (e.g., date formats, units like “kg” vs “kilograms”)

# Remove duplicate rows
df.drop_duplicates(inplace=True)

4. Feature Engineering

Not all features add value. Sometimes, combining or transforming features improves model performance.

Examples:

• Creating a new feature like BMI = weight / height²

• Dropping irrelevant features like “student ID”

• Selecting only features that affect predictions

    # Example: Create BMI from height & weight
    df['BMI'] = df['Weight'] / (df['Height']/100)**2
  
    # Drop irrelevant columns
    df.drop(columns=['Student_ID'], inplace=True)
  

  5. Encoding Categorical Data

Machine learning algorithms work with numbers, not text. So categorical data (like “Male/Female” or “Yes/No”) must be converted.

Common techniques:

• Label Encoding: Assigns numbers to categories (e.g., Male=0, Female=1)

• One-Hot Encoding: Creates separate binary columns for each category

• Ordinal Encoding: Useful when categories have an order (e.g., Small=1, Medium=2, Large=3)

6. Feature Scaling

Features like salary (in thousands) and age (in years) exist on different scales. Models may give more importance to larger values unless we scale them.

Popular techniques:

• Min-Max Scaling – converts values into a fixed range (e.g., 0–1)

• Standardization – rescales data to have mean=0 and standard deviation=1

Splitting Data into Train and Test Sets

Once preprocessing is complete, we split the dataset:

Training set (70–80%) – used to train the model

Testing set (20–30%) – used to evaluate performance

This ensures the model doesn’t just “memorize” data but generalizes well to unseen data.

from sklearn.model_selection import train_test_split

X = df.drop('Target', axis=1)  # features
y = df['Target']               # target variable

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.3, random_state=42
)

Target Variable vs Input Features

Target Variable (y): The outcome we want to predict

Example: Whether a patient has cancer (Yes/No)

Input Features (X): Independent variables used for prediction

Example: Age, blood pressure, cholesterol level

Final Thoughts

Data preprocessing may seem less exciting than building models, but it’s the most critical step in any ML pipeline. If you invest time in preparing high-quality data, your models will reward you with accuracy, stability, and reliability.