MLA-C01 Practice Question 81: Sensor Outlier Handling
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 9 · question 1 of 10
A factory is preparing temperature sensor data in Amazon SageMaker Canvas to train a predictive-maintenance model. The sensors are rated for -40°C to 125°C, but firmware glitches occasionally write placeholder readings of 9999 or -999. Each row also has vibration and pressure readings that the model needs. The engineer wants to get rid of the impossible temperatures without losing the rows, so that a later missing-value step can fill them in, and without altering any real readings, including legitimate extreme ones.
Which Handle outliers transform should the engineer apply to the temperature column?
Show answer and explanation
Answer:
The Min-max numeric outliers transform is meant for when you already know the thresholds that mark outliers, and the sensor's rated range supplies them exactly. Values above the upper or below the lower threshold count as outliers. The Invalidate fix method replaces those values with invalid values instead of dropping the row or changing them to another number. The glitch readings are removed, every row keeps its vibration and pressure data, and real readings inside the rated range are untouched (option C).
- A is wrong on two counts. Remove deletes whole rows, losing the other sensor readings, and values like 9999 inflate the mean and standard deviation, so the statistical bounds stop matching physical reality. - B is wrong because percentile bounds always cap the top and bottom 1% of values, including real extreme temperatures. Clipping also turns 9999 into a believable-looking reading instead of removing it. - D is wrong because Clip keeps the glitch rows but replaces 9999 with the upper bound, which invents a plausible temperature the sensor never measured.
Reference: Transform data - Handle Outliers (Amazon SageMaker Canvas)