NY City Restaurant Data Analysis

Analyze NY City's restaurants data, and find the following:

• Which factor(s) affect the number of days between restaurant's inspection?
• Which factor(s) affect the closure of a restaurant, in the current inspection?
• Which factor(s) affect the closure of a restaurant, in the next inspection?
• Can we predict the above three with some degree of certainty?

• Data about NY City restaurants ratings, inspection details and citations is publicly available at https://data.cityofnewyork.us/Health/DOHMH-New-York-City-Restaurant-Inspection-Results/xx67-kt59
• This data gets updated almost daily
• Lower the restaurant's score, better is the restaurant
• Each restaurant is inspected at least once an year and these happen suddenly, without any notice

• Data is loaded into normalized RDBMS tables. RDBMS tables help us to manage the data efficiently, and these can be updated daily with new data
• The data needed for analysis is obtained in the form of a CSV file(s) from RDBMS Tables
• The CSV files obtained from RDBMS are loaded at www.github.com. These files are accessed by R programs for data analysis
• Data is transformed, and two types of data sets are produced for training and model testing (…Cont)

• R language is used to transform the data to the required format (data frames)
• The model training data is taken from all the years data except the data from 2015 year
• The model testing data is taken from the 2015 year's data

• Found that the “score” variable alone shows strong relationship with the days between the successive inspections

\[ days=12.19-((2.859)(score^4)/(10^6)) \]

• Predicting if a restaurant will be closed, based on the current citations, score and other variables

LDA has better performance

• Predicting if a restaurant will be closed in the next visit, based on the current citations, score and other variables

QDA has better performance

• With more than 96 variables, and around 150 MB of data, the algorithms (developed in R) have ran while
• Creating the logic to get the days difference between consecutive inspections (using vectorized operations only)
• Deciding on which functional forms to use (parametric approach)
• kable function was not working properly at some places, and it took a while for me to figure out the problem
• Publishing the document in RPUBS also ran for a while, due to the volume of data we have

RPUBS at http://www.rpubs.com/msekhar12/MSDA_607_Final_Project

The source code and data files are present at the following location: https://github.com/msekhar12/MSDA_FINAL_PROJECT/tree/master

The first page's image (restaurant's ratings image) is found at google images