Professional Machine Learning Engineer

Question

You are developing a model to predict daily temperatures. You randomly split the data and then transformed the training and test datasets. Temperature data for model training is uploaded hourly. During testing, the model achieved 97% accuracy; however, after deployment to production, its accuracy fell to 66%. How can you improve the production model’s accuracy?

Explanation

Temperature observations are time-series data, so randomly splitting them can leak future or closely related observations into the training set and inflate offline accuracy. Splitting chronologically evaluates the model on later, unseen periods in the same way production does, yielding a more reliable model assessment and reducing temporal leakage.

Professional Machine Learning Engineer

AstroTutor

Notes

Readiness

QuestionQ57

Community Discussion