MACHINE LEARNING • LESSON 6

Incorrect Data

Incorrect data means that a value exists in the dataset, but the value itself is wrong, invalid, or does not represent the real-world situation correctly.

THE SIMPLEST DEFINITION

Incorrect data is information that is present but wrong.

This is different from a missing value. With missing data, we have no value. With incorrect data, we have a value, but that value is not correct.

01

Missing vs Incorrect Data

These two problems are easy to confuse.

MISSING DATA Age = ?

We don't know the person's age.

INCORRECT DATA Age = 250

A value exists, but it is probably wrong.

The key difference is simple:

Missing = information is unavailable. Incorrect = information is available but wrong.
02

Simple Example: Age

Suppose a customer dataset contains these ages:

25
31
28
250
34

An age of 250 is probably not a realistic customer age.

The problem is that the dataset contains a number, so the value is not technically "missing." The value itself is incorrect.

The model does not automatically know that 250 is a data-entry mistake.
03

How Incorrect Data Can Enter a Dataset

Incorrect values can come from normal human or system errors.

Typing Error Someone enters 250 instead of 25.
Wrong Unit A value may be recorded using the wrong unit.
System Error A software system may store an incorrect value.
Data Conversion Problem Converting data between systems can sometimes change values incorrectly.
04

Another Example: House Price

Imagine we are building a model to predict house prices.

HOUSE A $300,000
HOUSE B $450,000
HOUSE C $45,000,000

Suppose House C is an ordinary residential property and the intended price was actually $450,000.

The dataset now contains an incorrect value that is dramatically different from reality.

If we train a model without investigating the value, that incorrect information becomes part of the training data.

05

Incorrect Data Can Change What the Model Learns

A machine learning model learns patterns from the examples it receives.

If some examples contain incorrect information, the model can learn relationships that do not represent reality.

REAL DATA Correct values
PROBLEM Incorrect values
MODEL Learns from the data
RESULT Potentially poor predictions

This is why we should inspect suspicious values before using the dataset for training.

06

How Do We Handle Incorrect Data?

The correct action depends on why the value is incorrect and whether the real value can be determined.

Correct the Value If the original value is known, replace the incorrect value with the correct one.
Remove the Record If the record is unreliable and cannot be fixed, removing it may be appropriate.
Investigate Further Sometimes a surprising value is actually valid, so don't delete it just because it looks unusual.
07

Real-World Example

Imagine an e-commerce company has customer ages:

Customer Age Orders Status
Rahul 29 5 Valid
Priya 290 3 Check
Arun 34 8 Valid

We should not immediately tell the model that Priya is 290 years old.

Instead, we investigate the source. Perhaps the actual value was 29 and someone entered an extra zero.

DATASET Age = 290
AFTER VERIFICATION Age = 29

Now the dataset contains the verified value instead of the incorrect value.

IMPORTANT

Don't Confuse Incorrect Data With Unusual Data

A value that looks unusual is not automatically incorrect.

For example, suppose a dataset contains a person's annual income of ₹5 crore. That value may look unusual compared with most customers, but it could be completely real.

Unusual does not mean incorrect. Verify the data before changing or deleting it.

This distinction becomes especially important when we learn about outliers in the next topic.

REMEMBER THIS

Don't Let Wrong Information Become Training Data.

Incorrect data contains a value, but that value does not correctly represent reality. Find suspicious values, investigate them, and correct or remove them when there is enough evidence that they are wrong.

QUICK CHECK

Which One Is Incorrect Data?

Age = ? Missing data
Age = 250 Likely incorrect data
Age = 95 Could be valid — investigate first
Answer

Age = 250 is likely incorrect for a normal customer dataset, but it should still be investigated. Age = 95 is unusual but could be completely valid. Never assume that an unusual value is automatically wrong.

NEXT TOPIC

Outliers

Some values are very different from most other values. Next, we will learn what these unusual values are called, why they happen, and when they should or should not be removed.