Why data preparation consumes most ML time in ML Python - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why does data cleaning take so much time in ML projects?

In machine learning, data cleaning is a major part of data preparation. Why does it usually take the most time?

A Because raw data often has errors, missing values, and inconsistencies that need fixing before training.

B Because training models requires a lot of computing power and time.

C Because data cleaning involves writing complex algorithms to improve model accuracy.

D Because data cleaning is automated and runs slowly on large datasets.

🧠 Conceptual

intermediate

What is the main reason feature engineering is time-consuming?

Feature engineering is a key step in data preparation. Why does it often consume a lot of time?

A Because it only works on small datasets and needs manual scaling.

B Because it involves training multiple models to select features.

C Because it requires domain knowledge to create meaningful features from raw data.

D Because it is fully automated and requires tuning many parameters.

❓ Metrics

advanced

Measuring data quality impact on model accuracy

You have two datasets: Dataset A is clean and Dataset B has 20% missing values. You train the same model on both. Which metric difference best shows the impact of data quality?

A Dataset A precision: 0.5, Dataset B precision: 0.5

B Dataset A loss: 0.2, Dataset B loss: 0.2

C Dataset A training time: 10 minutes, Dataset B training time: 15 minutes

D Dataset A accuracy: 90%, Dataset B accuracy: 70%

🔧 Debug

advanced

Why does this data scaling code cause poor model results?

Consider this Python snippet for scaling features before training:

from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.fit_transform(X_test)

Why might this cause poor model performance?

A Because the scaler is fit separately on test data, causing data leakage and inconsistent scaling.

B Because StandardScaler cannot be used on numeric data.

C Because scaling is not needed before training models.

D Because fit_transform should be replaced with transform on training data.

❓ Model Choice

expert

Choosing the best approach to handle missing data in a large dataset

You have a large dataset with 30% missing values scattered randomly. Which approach is best to prepare data for a machine learning model?

A Remove all rows with missing values to keep only complete data.

B Fill missing values with the mean or median of each feature before training.

C Use a model that can handle missing values natively without imputation.

D Replace missing values with zeros regardless of feature meaning.