Assignment A3.1 - Predictive Modeling of House Prices

Assignment Description Please note: this is an individual assignment. For the final course project, students should create a single replicable notebook building a ML model for predicting house prices in the test dataset located in the file “[login to view URL]” by training the model in the train dataset that can be found in the file “[login to view URL]”. In order to earn points, students must complete the tasks (listed below) and also explain it in the comments. You can reuse your own work from the activities and the Week 2 assignment. Steps to Follow Follow these steps to complete the assignment: 1. Train-test separation and any other subset needed (encoding, validation) 2. Variable transformation in both train and test: Apply one-hot encoding to at least one variable Create a new variable from scratch and explain why you think it can be useful Apply target encoding to at least one variable Correctly apply one of the algorithms seen in the course 3. Choosing two parameters (at least) to tune: Explain the intuition: “How is the algorithm affected by the parameter being higher or lower?” Explain how you chose the final values for that parameter Explain the error metric used for performance assessment and the values you obtained in the different subsets Apply your model to the test dataset and measure your error 4. Identify the top five most important variables in your model and provide your own interpretation: Why are they important, and how do they impact your prediction? (“Is it intuitive or counterintuitive? Is the predicted value higher or lower when this variable’s value increases?”) 5. If possible, add some external data to solve this problem: what other factors not included in this dataset can help to predict our target? Submission The assignment must be delivered in a Python notebook. 1) Please follow the attached files 2) No AI 3) Follow the comments from pervious assignment A2.1 Comment: Hello Mohammed, Thank you for submitting Assignment A2.1. Here is your feedback: Your submission demonstrates exceptional technical sophistication, particularly in your implementation of leakage-safe target encoding using K-fold cross-validation. This advanced technique shows you understand not just how to create features, but how to avoid common pitfalls that compromise model validity. Your three engineered features-the square root transformation capturing diminishing returns, the quality-area interaction term, and the neighborhood encoding-are genuinely creative and well-justified with clear business logic. The code is professionally structured with comprehensive comments, fixed random seeds for reproducibility, and clear visualizations that effectively communicate your findings. While your technical execution is strong,the submission would benefit from deeper initial data exploration to better understand feature distributions and relationships before engineering. Your model interpretation correctly identifies the top features but could provide richer business insights-for instance, explaining why neighborhood encoding ranks second might reveal interesting market dynamics about location premiums. The relatively low importance of Sart_LivArea (0.0107) compared to your other features suggests this transformation might need refinement or that the relationship isn't as non-linear as hypothesized. Consider these enhancements for your next assignment: Add exploratory visualizations (histograms, scatter plots, correlation matrices) during initial data understanding to guide more targeted feature engineering decisions When interpreting feature importance. connect findings back to real-world business implications-what does it mean for buyers, sellers, or investors that the quality-area interaction dominates predictions? Experiment with alternative non-linear transformations or polynomial features that might better capture the complexity you're targeting, especially given the weak performance of your square root transformation Rubric Your work shows strong technical skills and creative thinking. Keep pushing yourself to connect the technical aspects more deeply with domain knowledge and business value. Best regards,

Project ID: 40539998