Design a house-price prediction workflow
Overview
This question evaluates end-to-end machine learning system design competencies, including data preprocessing, feature engineering, model selection, evaluation, monitoring, interpretability, fairness, and deployment for price prediction, and is categorized under Machine Learning with a focus on practical application complemented by conceptual understanding. It is commonly asked to assess an engineer's system-level thinking in translating business requirements into a robust ML workflow that handles data quality, non-stationarity, evaluation trade-offs, and latency/memory constraints.
Predicting Home Sale Prices: End-to-End ML Design
Context
You have historical home-sale records with features such as lot area, year built, number of rooms, neighborhood, and sale date. You need to build a production-ready system that predicts sale price for new listings.
Task
Design an end-to-end approach that covers:
- Problem framing and target transformation (e.g., log-price) and objective.
- Data cleaning: missing-value imputation, outlier handling, and leakage risks.
- Feature engineering for numeric, categorical, time, and location features; useful interactions.
- Baselines and model choices: linear models, regularized regressions, tree-based methods, ensembles.
- Evaluation protocol and metrics: RMSE, MAPE, and time-based cross-validation.
- Hyperparameter tuning and error analysis.
- Interpretability and fairness considerations.
- Handling non-stationarity and market shifts: monitoring and retraining.
- Deployment plan for a lightweight model under latency and memory constraints.
Solution
Login required