Predict Future Car Sale Prices
Problem Description
You are provided with transaction-level data representing completed car sales from various dealers throughout the calendar year 2018. Your primary objective is to develop a predictive model capable of estimating future vehicle sale prices for transactions occurring after 2018.
Each row in the dataset corresponds to a single completed vehicle sale and includes the following fields:
sale_date: The date of the transaction.dealer_id: Unique identifier for the selling dealer.make,model,trim,model_year: Specific vehicle identifiers.mileage: Odometer reading at the time of sale.condition_score: A numerical score indicating the vehicle's condition.fuel_type,transmission,body_type,color: Vehicle specifications.accident_history: Indicator of past accidents.number_of_owners: Count of previous owners.dealer_location: Geographical location of the dealer.final_sale_price: The actual price at which the vehicle was sold (this is your target variable).
You are required to address the following specific questions:
- Problem Framing and Data Strategy: How would you frame this as a prediction problem? Define the target variable. Crucially, how would you split the data to ensure that your model's evaluation accurately reflects its performance on future, unseen transactions, rather than merely fitting historical data randomly?
- Feature Engineering: Describe the feature engineering steps you would undertake to prepare the vehicle price prediction model.
- Feature Exclusion: Identify which features from the provided dataset you would exclude. Justify your exclusions by explaining how they might lead to target leakage or why they would not be available at the time of prediction.
- Model Selection and Evaluation: Which types of models and evaluation metrics would you compare for this task? How would you specifically address challenges such as nonlinearity in relationships, seasonality in sales, and potential distribution shifts between the 2018 training data and future prediction periods?
- Linear Regression Interpretation: Imagine you are presented with the output of a fitted linear regression model, including coefficients, standard errors, p-values, and R-squared. How would you interpret the coefficients for both continuous and categorical variables? What important caveats would you raise before using these results to inform business decisions?
- Selection Bias: Given that the dataset exclusively contains completed sales and does not include information on unsold inventory, what potential selection-bias issues might arise, and how could they impact your model?
AI interview prep
Core dilemma
Predicting future values in a dynamic market using static historical data, specifically managing temporal dependencies, preventing data leakage, and ensuring model robustness against distribution shifts over time.
3 tradeoffs and 3 stress tests locked
Interview prep
Unlock the interviewer X-Ray rubric, tradeoff matrix, stress tests, clarifying questions, and key concepts.
Predict Future Car Sale Prices
Problem Description
You are provided with transaction-level data representing completed car sales from various dealers throughout the calendar year 2018. Your primary objective is to develop a predictive model capable of estimating future vehicle sale prices for transactions occurring after 2018.
Each row in the dataset corresponds to a single completed vehicle sale and includes the following fields:
sale_date: The date of the transaction.dealer_id: Unique identifier for the selling dealer.make,model,trim,model_year: Specific vehicle identifiers.mileage: Odometer reading at the time of sale.condition_score: A numerical score indicating the vehicle's condition.fuel_type,transmission,body_type,color: Vehicle specifications.accident_history: Indicator of past accidents.number_of_owners: Count of previous owners.dealer_location: Geographical location of the dealer.final_sale_price: The actual price at which the vehicle was sold (this is your target variable).
You are required to address the following specific questions:
- Problem Framing and Data Strategy: How would you frame this as a prediction problem? Define the target variable. Crucially, how would you split the data to ensure that your model's evaluation accurately reflects its performance on future, unseen transactions, rather than merely fitting historical data randomly?
- Feature Engineering: Describe the feature engineering steps you would undertake to prepare the vehicle price prediction model.
- Feature Exclusion: Identify which features from the provided dataset you would exclude. Justify your exclusions by explaining how they might lead to target leakage or why they would not be available at the time of prediction.
- Model Selection and Evaluation: Which types of models and evaluation metrics would you compare for this task? How would you specifically address challenges such as nonlinearity in relationships, seasonality in sales, and potential distribution shifts between the 2018 training data and future prediction periods?
- Linear Regression Interpretation: Imagine you are presented with the output of a fitted linear regression model, including coefficients, standard errors, p-values, and R-squared. How would you interpret the coefficients for both continuous and categorical variables? What important caveats would you raise before using these results to inform business decisions?
- Selection Bias: Given that the dataset exclusively contains completed sales and does not include information on unsold inventory, what potential selection-bias issues might arise, and how could they impact your model?