AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Encrypting sensitive data before storing it in Azure for security compliance

  Why it's wrong here

  Encrypting sensitive data is a security control that converts plaintext into ciphertext to protect confidentiality, but it does not alter the statistical or semantic content that an ML model needs. Preprocessing, in contrast, transforms raw values into a clean, consistent representation—imputing missing entries, normalizing ranges, and encoding categories—so algorithms can interpret them correctly. Encrypted data would actually be unusable for training unless decrypted first, and it has no effect on issues like nulls or scale.

• ✓

  Transforming raw data (handling nulls, scaling, encoding) to make it suitable for ML training

  Why this is correct

  Data preprocessing is the foundational transformation that converts raw, messy data into a clean, structured form for ML training: missing values are imputed or dropped, numerical features are scaled (e.g., Min-Max, z-score normalization) to prevent one feature from dominating, and categorical labels are encoded (one-hot, ordinal) into numeric vectors. It directly impacts model accuracy and training stability, as algorithms like gradient descent converge faster with normalized inputs and tree models require encoded categories. Without preprocessing, nulls can crash algorithms, wide-scale differences bias distance-based models, and string features are simply incompatible.

• ✗

  The process of splitting raw data into training and test sets

  Why it's wrong here

  Splitting data into training and test sets is a model evaluation technique that partitions a dataset to estimate generalization and detect overfitting, not a transformation of the feature data itself. Unlike preprocessing—where missing values are imputed, features are scaled, and categorical text is converted to numeric codes—splitting simply assigns existing rows to different groups. Moreover, splitting must happen before any fitting of imputers/scalers to avoid data leakage, which makes it a workflow step distinct from data transformation.

• ✗

  Compressing data files to reduce the cost of Azure Blob Storage

  Why it's wrong here

  Compressing data files is a storage optimization that reduces byte size using algorithms like gzip or snappy, lowering Azure Blob Storage costs and transfer times, but the decompressed content remains the same raw data. Preprocessing is a machine-learning-specific step that changes the actual values and quality of the dataset—filling in nulls, removing outliers, and standardizing ranges—so the model can learn meaningful patterns. Compression doesn't cure missing values, categorical variables, or feature scale mismatches, so it is unrelated to ML readiness.

Go deeper

Related to this question