What are the Techniques for Handling Missing Data in Machine Learning?

What are the Techniques for Handling Missing Data in Machine Learning?

In the world of machine learning, data is king. The more data you have, the better you can train your models to make accurate predictions. But what happens when your data is riddled with missing values and inconsistencies? Missing data can be a real challenge for machine learning practitioners, but fear not! This blog equips you with an arsenal of techniques to address and in handling missing data and get your models back on track.

What is a Missing Value?

A missing value, also known as missing data, represents an absent data point within a dataset. These absences can arise for various reasons, including human error during data collection, equipment malfunctions, or sensor failures.

Types of Missing Data

Types of Missing Data

Missing Completely at Random (MCAR)

If the missing data occurs randomly and is unrelated to other variables, techniques like mean/median/mode imputation can be effective.

Missing At Random (MAR)

If the missing data is related to observed variables but not missing values, imputation techniques like KNN or model-based imputation can be appropriate.

Missing Not At Random (MNAR)

This scenario presents a particular challenge for handling missing values. In this scenario, the missing data is related to missing values, making it more challenging to attribute accurately. Techniques like model-based imputation that account for the missingness mechanism might be necessary.

A Data Science Course in Chennai can be a valuable investment if you’re looking to acquire a comprehensive understanding of data science concepts, including handling missing data. You’ll learn not only about missing data imputation techniques for missing data but also about data cleaning, feature engineering, model building, and more.

Handling Missing data in machine learning

Some models are less sensitive to missing data than others. Decision trees can handle missing data implicitly, while models like linear regression might be more sensitive and require specific imputation techniques.

Missing data introduces hurdles for machine learning models in several ways.

  • Firstly, it reduces the volume of data available for training, potentially leading to less accurate models.
  • Secondly, it can introduce bias if the missing data isn’t random.

For instance, if customers with lower incomes are more likely to have missing income data, a model trained on such data might underestimate the income of future customers.

Techniques for Handling Missing Values

Techniques for Handling Missing Values

The optimal technique to address missing data in machine learning hinges on the specific characteristics of your data and the machine learning model you’re employing. Here’s a breakdown of some widely used techniques:

  • Deletion

The most straightforward approach to handling missing values involves eliminating rows or columns containing missing values. However, this method can be hazardous and result in information loss. Deletion is only recommended when the amount of missing data is minimal, and the missing data is distributed randomly.

  • Imputation

Imputation is the process of filling in and handling missing data with estimated values. Here are a few standard imputation techniques:

  • Mean/Median/Mode Imputation

Mean/Median/Mode imputation is a straightforward approach for handling missing data in machine learning. This technique replaces missing values with a statistic calculated from the existing data points within a feature.

This technique replaces missing values with the respective feature’s mean, median, or mode.

While it’s simple and easy to implement, it can be biased if the data isn’t normally distributed.

  • Mean Imputation

This method replaces missing values with the average value of the feature. For instance, if you have a feature representing customer age and some values are missing, you can replace those missing values with the average age of all customers in your dataset.

  • Median Imputation

This technique tackles missing values in ordered features by replacing them with the middle value, also known as the median. In handling missing values, median imputation is particularly useful for numerical features with a precise order, such as income, age, or customer service wait times.

  • Mode Imputation 

This method replaces missing values with the most frequent value for the feature. For instance, if you have a feature representing the country of residence and some values are missing, you can replace those missing values with the most frequent country in your dataset.

  • K-Nearest Neighbors (KNN) Imputation

This technique leverages the values of the nearest neighbours to estimate the missing value. KNN imputation can be more accurate than mean/median/mode imputation but comes at a higher computational cost.

Here’s a breakdown of how KNN imputation works for handling missing values:

  • Define the value of k, which represents the number of nearest neighbours to consider.
  • For each data point with a missing value, identify the k nearest neighbours based on the available features.
  • To estimate the missing value, use a voting mechanism (for categorical data) or calculate the average (for continuous data) of the k nearest neighbours’ values.

For instance, consider you have a dataset with features like age, income, and credit score. If a data point has a missing value for income, KNN imputation would identify the k closest data points based on age and credit score and then use the average income of those neighbours to estimate the missing value.

By leveraging the relationships between similar data points, KNN imputation offers a more sophisticated approach to handling missing data in machine learning than simpler methods like mean/median/mode imputation.

  • Model-Based Imputation

This technique employs a machine learning model to predict the missing values. It can be highly accurate but can also be more complex to implement compared to other imputation techniques. Here’s a more in-depth look at model-based imputation:

  1. Choose a machine learning model, such as linear regression, decision tree, or random forest.
  2. Train the model on the data portion where there are no missing values. The target variable for the model will be the feature for handling missing values you want to blame.
  3. Use the trained model to predict the missing values in the features with missing data.

By understanding these techniques and their nuances, you can develop a data pre-processing strategy that effectively addresses missing data in your machine-learning projects. Consider pursuing a Data Science Course in Bangalore to delve deeper into these concepts and gain hands-on experience handling missing data using popular Python libraries like Pandas and sci-kit-learn.

Additional Considerations

Domain Knowledge 

Incorporate domain knowledge about your data when selecting a missing data technique. For instance, if you know that income data is typically left blank for low-income customers, KNN imputation might be more appropriate than mean imputation.

Iterative Approach

Experiment with various techniques and evaluate their impact on your model’s performance. You might find that a combination of techniques works best for your specific dataset.

Handling Missing Data for Categorical Variables

Unlike numerical data, where techniques like mean or median imputation can be applied, categorical data presents unique challenges when it comes to imputing missing values. Here’s why:

Meaningless Imputation

Simply replacing missing values with the mean or median for categorical data doesn’t make sense. For example, if the feature is “colour” and a data point has a missing value, replacing it with the average (“blue” + “red” + “green”) / 3 wouldn’t be a valid colour.

Loss of Information

Specific techniques employed for handling missing data in categorical variables can introduce information loss. A prime example is mode imputation, which replaces missing values with the most frequent category.

Alternative Imputation Techniques for Categorical Data

Mode Imputation

This option remains viable for categorical data, replacing missing values with the most frequent category.

Category-Specific Imputation

This technique tackles the challenge of handling missing values in categorical data by leveraging the inherent structure within categories. Assign missing values within a category based on the distribution of other observed values within that category. For instance, if the feature is “colour” and “red” is a missing value for a particular data point, you could analyse the category “red” and impute the missing value with the most frequent sub-category (e.g., “bright red” or “dark red”).

Proximity-Based Techniques

Utilise techniques like KNN imputation but consider distance metrics suitable for categorical data. This might involve measuring similarity based on co-occurrence with other categories or shared properties.

Choosing the Right Technique for Handling Missing Data

The optimal technique for handling missing data depends on several factors specific to your data and model:

Amount of Missing Data

Deletion might be a viable option if the amount of missing data is minimal. However, imputation techniques for missing data become more suitable for larger amounts of missing data.

Evaluating the Impact of Missing Data Techniques

It’s crucial to assess the impact of your chosen technique on your machine learning model’s performance. Here are some evaluation methods:


Cross-validation is a powerful tool to assess the effectiveness of different strategies for handling missing values. Split your data into training and testing sets. Apply different missing data techniques to the training set and evaluate the model’s performance on the held-out testing set. Choose the technique that yields the best performance.

Metric Comparison

Compare your model’s performance with different missing data handling techniques using relevant metrics like accuracy, precision, recall, or F1 Score.

Visualisation Techniques 

Techniques like boxplots or violin plots can help visualise the distribution of data before and after imputation, allowing you to assess the effectiveness of the imputation method.

Encoding & Handling Missing Values as a Feature

Another approach for handling missing data in machine learning involves creating new features to indicate the presence of missing values. This technique can benefit certain machine learning models, like decision trees, that can effectively handle categorical features.

Here’s a breakdown of two common encoding methods:

Binary Encoding

Introduce a new binary feature (e.g., “has_missing_age”) where 1 indicates a missing value for age and 0 represents a valid value. This approach is suitable for both numerical and categorical features with missing values.

One-Hot Encoding

This technique is used for handling missing data. It creates a new categorical feature for each possible value of the original feature, including a category for missing values.

For instance, if the original feature is “country” with values like “USA,” “Canada,” and “UK,” one-hot encoding would create three new features: “is_USA,” “is_Canada,” and “is_UK.” A data point with a missing value in the original “country” feature would have all three new features set to 0. One-hot encoding is recommended for categorical features with a manageable number of categories.

Ready to Master Data Analysis and Conquer Missing Data Challenges? Enroll in our comprehensive Data Analytics Course in Chennai and gain the skills and knowledge to effectively handle missing data in your machine learning projects.

What is SVM?

Support Vector Machine (SVM) is a robust supervised learning algorithm that excels at classification tasks, even when dealing with challenges like handling missing data in machine learning. Imagine you have a dataset where you want to classify data points into distinct categories, like classifying emails as spam or not. SVMs achieve this by creating a hyperplane, essentially a decision boundary in high-dimensional space, that separates the different categories with the maximum possible margin.

SVM handle missing data in machine learning with the following working components:

Feature Space

Data points are mapped to a high-dimensional feature space, where features might be individual attributes or engineered combinations of features. This mapping allows SVMs to handle complex relationships between features, which is particularly beneficial for non-linear data.

Support Vectors

The data points closest to the hyperplane on either side are called support vectors. These points are crucial in defining the optimal hyperplane that increases the margin between the classes.

Maximising the Margin

The core objective of SVMs is to find the hyperplane that separates the classes with the most significant margin. This margin refers to the distance between the hyperplane and the closest support vectors from each class. A larger margin translates to a more robust classification model, less prone to errors on unseen data.

By understanding SVMs and their ability to handle missing data, you can unlock their potential for various classification tasks. Consider supplementing your knowledge with Machine Learning Training in Chennai to delve deeper into this field and explore the vast applications of machine learning.

Can SVM Handle Missing Data?

Having explored various techniques for handling missing data, let’s address a common question: how do SVMs handle missing information? While SVMs are powerful machine learning models, they aren’t explicitly designed to accommodate missing data. Here’s a breakdown of their interaction with missing values:

Partial Tolerance

Unlike some models that require complete data points for training, SVMs can function with a certain degree of missing data. This is because they primarily focus on a subset of data points closest to the decision boundary, which might not always have missing values.

Sensitivity to Missing Data Type

The impact of missing data on SVMs can vary depending on the type of kernel used. Linear SVMs might be more sensitive to missing data than non-linear SVMs with kernels like the radial basis function (RBF kernel).

To ensure optimal performance from your SVM model, it’s strongly recommended to address missing data using the techniques mentioned earlier (deletion, imputation, etc.) as part of your handling missing data strategy.

By pre-processing your data to address missing information, you can mitigate the adverse effects and effectively leverage SVMs for machine learning tasks.

If you want to know more about the latest salary trends for Data Scienctist, Check out Data Scientist Salary For Freshers, which will help you get an insight into the packages as per the companies, skills and experience.

Handling Missing Data with Graph Representation Learning

While this blog focuses on more traditional techniques, it’s essential to acknowledge the existence of advanced approaches:

Graph Representation Learning

This approach represents data points as nodes in a graph and connections between data points as edges. By leveraging the relationships and properties of connected nodes, missing data can be handled.

Potential Benefits

Graph representation learning can potentially capture complex relationships between data points and infer missing values based on these relationships. However, it’s a more advanced technique requiring domain knowledge and expertise in graph algorithms.

Handling Missing Data in Pandas

Pandas, a popular Python library for data analysis and manipulation, offers functionalities to address missing data:

Identifying Missing Data

Use functions like isnull() and isna() to identify missing values in your pandas DataFrame. These functions return a DataFrame with boolean values indicating missing entries.

Deletion Techniques

  • dropna(): This function drops rows or columns containing missing values. You can specify which axis (rows or columns) and a threshold for the minimum number of non-missing values per row/column to keep.
  • drop(axis=0, inplace=True): This allows you to drop rows with missing values permanently by setting inplace=True.

Imputation Techniques

  • fillna(): This function allows you to fill missing values with a specific value (e.g., the mean, median, or a custom value).
  • interpolate(): This function interpolates missing values for numerical data using techniques like linear interpolation or spline interpolation.

Example Code Snippet:


import pandas as pd

# Sample data with missing values

data = {'Age': [25, None, 30, None], 'Income': [50000, None, 70000, 40000]}

df = pd.DataFrame(data)

# Identify missing values


# Drop rows with missing values (axis=0 for rows)

df_dropna = df.dropna(axis=0)

# Fill missing values in 'Age' with the mean

df_filled = df.fillna(method='ffill', limit=1) # Fill with previous value (ffill)

# Print results



This blog has equipped you with a foundational understanding of handling missing data in machine learning. As you delve deeper into this field, explore advanced techniques like graph representation learning and stay updated on the evolving landscape of data manipulation tools. To further solidify your knowledge of handling missing data, consider enrolling in FITA Academy‘s data manipulation and machine learning courses to hone your skills further!

Quick Enquiry

Please wait while submission in progress...

Contact Us


  93450 45466


 93450 45466


 95978 88270


93450 45466


97900 94102


93635 21112

For Hiring

 93840 47472

Corporate Training

 90036 23340

Read More Read less

FITA Academy Branches




Other Locations

FITA Academy - Velachery
Plot No 7, 2nd floor,
Vadivelan Nagar,
Velachery Main Road,
Velachery, Chennai - 600042
Tamil Nadu

    :   93450 45466

FITA Academy - Anna Nagar
No 14, Block No, 338, 2nd Ave,
Anna Nagar,
Chennai 600 040, Tamil Nadu
Next to Santhosh Super Market

    :   93450 45466

FITA Academy - T Nagar
05, 5th Floor, Challa Mall,
T Nagar,
Chennai 600 017, Tamil Nadu
Opposite to Pondy Bazaar Globus

    :   93450 45466

FITA Academy - Tambaram 27, Nehru Nagar, Kadaperi, GST Road, West Tambaram, Chennai 600 045, Tamil Nadu Opposite to Saravana Jewellers Near MEPZ

    :   93450 45466

FITA Academy - Thoraipakkam
5/350, Old Mahabalipuram Road,
Okkiyam Thoraipakkam,
Chennai 600 097, Tamil Nadu
Next to Cognizant Thoraipakkam Office and Opposite to Nilgris Supermarket

    :   93450 45466

FITA Academy - Porur
17, Trunk Rd,
Chennai 600116, Tamil Nadu
Above Maharashtra Bank

    :   93450 45466

FITA Academy Marathahalli
No 7, J J Complex,
ITPB Road, Aswath Nagar,
Marathahalli Post,
Bengaluru 560037

    :   93450 45466

FITA Academy - Saravanampatty
First Floor, Promenade Tower,
171/2A, Sathy Road, Saravanampatty,
Coimbatore - 641035
Tamil Nadu

    :   95978 88270

FITA Academy - Singanallur
348/1, Kamaraj Road,
Varadharajapuram, Singanallur,
Coimbatore - 641015
Tamil Nadu

    :   95978 88270

FITA Academy - Madurai
No.2A, Sivanandha salai,
Arapalayam Cross Road,
Ponnagaram Colony,
Madurai - 625016, Tamil Nadu

    :   97900 94102

FITA Academy - Pondicherry
410, Villianur Main Rd,
Sithananda Nagar, Nellitope,
Puducherry - 605005
Near IG Square

    :   93635 21112

Read More Read less
  • Are You Located in Any of these Areas

    Adyar, Adambakkam, Anna Salai, Ambattur, Ashok Nagar, Aminjikarai, Anna Nagar, Besant Nagar, Chromepet, Choolaimedu, Guindy, Egmore, K.K. Nagar, Kodambakkam, Koyambedu, Ekkattuthangal, Kilpauk, Meenambakkam, Medavakkam, Nandanam, Nungambakkam, Madipakkam, Teynampet, Nanganallur, Navalur, Mylapore, Pallavaram, Purasaiwakkam, OMR, Porur, Pallikaranai, Poonamallee, Perambur, Saidapet, Siruseri, St.Thomas Mount, Perungudi, T.Nagar, Sholinganallur, Triplicane, Thoraipakkam, Tambaram, Vadapalani, Valasaravakkam, Villivakkam, Thiruvanmiyur, West Mambalam, Velachery and Virugambakkam.

    FITA Velachery or T Nagar or Thoraipakkam OMR or Anna Nagar or Tambaram or Porur branch is just few kilometre away from your location. If you need the best training in Chennai, driving a couple of extra kilometres is worth it!