Midterm Peer Review

[bhyeh]

Project Summary

The topic of this project is on earthquakes in Nepal. The data used is survey information from two sources: [1] a lab [2] Nepal Bureau of Statistics. The project aims to predict the severity of damage to buildings caused by earthquakes in Nepal.

Three Things I Like

• The actual flow of the report itself is very concise and each step of the project is succinctly described and reasoned very well
• I like how with your choice of standardizing you provide clear reasoning to why you are performing standardization, and also acknowledge the reason for a different standardization
• Again, the flow of scatter and correlation plots runs great alongside the deep dive into your different features - it is very easy to understand each step of your data process

Three Areas for Improvement

• On the topic of using ensemble trees, it might be interesting to use the features found by a Boosting ensemble in another model(OLS nominal, etc.)
• The exploratory analysis into the features present in the data frame is very sufficient, but there was no mention to possible feature engineering of new features. If it not suitable, maybe providing reason to why the group thought feature engineering would be inappropriate would be nice
• Are there other external factors that may influence the damage grade of a building caused by an earthquake? is there any information to where the buildings are located? For e.g., does being closer/further from a fault line/valley/mountainous reigon/etc. influence how damaged a building is?

[YoannaEf]

We don't have a lot of information about the areas in which the buildings are located. However, there are three integer-valued variables (geo_level_1_id, geo_level_2_id, geo_level_3_id) in the dataset that correspond to the geographic region in which the building exists. If we want to see if a building is closer/further from a valley, mountain, etc. we will have to look at how those three variables were created and which exact area in Nepal corresponds to each number.