AI0-001 AI Models and Data Engineering Practice Question

Answer analysis

Option-by-option breakdown

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

• ✗

  Replace missing sensor values with the overall mean of the entire dataset.

  Why it's wrong here

  Imputing with the global mean ignores temporal locality and can introduce bias, especially if missingness is related to failure events. Time-series data often requires forward-fill or interpolation to preserve trends. Global mean imputation can mask anomalies and degrade model performance. It is not a feature engineering technique for capturing temporal patterns.

• ✗

  Apply one-hot encoding to the timestamp column to represent each time point as a binary vector.

  Why it's wrong here

  One-hot encoding timestamps creates high-dimensional, sparse features and does not capture temporal relationships. It treats each time point independently, losing sequential information. This is inappropriate for time-series data where trends and cycles matter. It would also lead to overfitting and poor generalization.

• ✓

  Compute rolling window statistics such as mean, standard deviation, and min/max over recent time intervals.

  Why this is correct

  Rolling window statistics capture temporal patterns and trends, such as increasing variance before failure. They summarize recent behavior and are effective features for anomaly detection in sensor data. These features help models identify deviations from normal operating conditions, improving predictive performance.

• ✗

  Perform principal component analysis (PCA) on the raw sensor readings to reduce dimensionality.

  Why it's wrong here

  PCA can reduce dimensionality but may destroy temporal structure and interpretability. It is not specifically suited for time-series feature engineering. While it can be applied, it does not capture the sequential dependencies needed for anomaly detection. Rolling statistics and lag features are more directly relevant and effective for this task.

• ✓

  Extract lag features by including previous sensor readings as additional input variables.

  Why this is correct

  Lag features incorporate past values, allowing the model to learn temporal dependencies and detect changes over time. They are fundamental in time-series forecasting and anomaly detection. By including lags, the model can recognize patterns that precede failures, such as gradual increases in temperature or vibration.