The project aims to analyze the relationship between social and spatial factors and provide insight into the intangible and physical organization of the city through the perspective of dogs and their daily routines. It will explore socio-economic indicators such as property values, neighborhood quality, proximity to open spaces like parks, and walkability of the urban environment to visualize the city's spatial characteristics.

it all started from here...

Our project is motivated by the prevalence of certain dog breeds in various neighborhoods in Cambridge and Boston, and we seek to explore the correlation between dogs and socioeconomic factors such as income, work-life balance, household size, and neighborhood quality. While we would ideally like to examine individual dog and owner characteristics, ethical data inquiry practices prevent us from doing so, and data limitations make it impractical. Instead, we will use zip code as a common feature in our data collection and analysis. To conduct our investigation, we have chosen New York City as our research site due to the availability of comprehensive data on dogs, people, and neighborhoods, as well as its socioeconomic stratification and spatial segregation. Additionally, New York City is renowned as a bustling metropolis, making it an ideal location to investigate the relationship between dogs and urban living.

about the time series...

The time series plot reveals several patterns in the data on dog registrations. Firstly, the number of registrations remains relatively stable with a slight upward trend prior to the year 2020. Secondly, there are seasonal fluctuations within each year, with registration numbers increasing until September before declining. Finally, after the COVID-19 pandemic lockdown began in March 2020, registration numbers began to grow at a much faster rate than before. This might be due to social isolation and seeking comfort and companionship during the pandemic.

what happened to dog population...

Upon conducting an initial analysis of the data, we have discovered that Manhattan, Brooklyn, and Queens have the highest number of registered dogs, whereas the Bronx and Staten Island have notably fewer. Upon further investigation, we compared the number of registered dogs to the population in each borough and found that while the Bronx has a significantly lower number of registered dogs, it only has a 6% population difference compared to Manhattan (with a population of 1,471,160 in the Bronx and 1,664,727 in Manhattan).

what are the correlations...

To visualize the correlation between our datasets and dog information, we have categorized our datasets and compared them. Our correlation matrix has enabled us to quickly understand our datasets and identify the main factors that influence the dog population. We have found that factors such as park availability, commute time, income, and household size have the strongest correlation with the number of registered dogs.

Our datasets are organized into the following categories:
- Age, race, sex, population, income, poverty
- Citizenship, language
- Occupation, worker status, time to go to work, commute time, college
- Housing, vehicle, household, park

looking at the map...

Upon examining the map, we have discovered that there is not a direct correlation between the population of a given zip code region and the number of registered dogs in that area. For instance, although Brooklyn is more densely populated than some other areas, it has fewer registered dogs. This finding is not unexpected, given that zoning and land use, as well as various socioeconomic factors, can differ across neighborhoods. We plan to further explore these factors in our analysis. On the other hand, we have found that the Manhattan region has a higher number of registered dogs, particularly in neighborhoods near Central Park. It is worth noting that these areas tend to have higher rent prices and shorter commute times. From feature importance of Decision Tree, Gradient Boosting, we see several features that are highly relevant to the dog counts in each Zip Code: Park Acres, Park Count, Population, Dog Number

what are the factors...

We have identified several factors that are strongly correlated with the number of registered dogs in different neighborhoods of New York City. These factors include park acres, commute time, and median income.

Park Acres: Parks are popular destinations for dog owners and their pets. As a result, areas with more parkland tend to have higher numbers of registered dogs. Additionally, parks are often located near residential areas and less frequently found in industrial areas.

Commute Time: Workers in the Manhattan region have shorter commutes compared to workers in other boroughs, with an average commute time of 32 minutes. Our analysis has shown that the number of registered dogs is significantly higher in the area around Manhattan. This is likely due to the higher density of jobs in Manhattan, as well as its association with higher socioeconomic status.

Median Income: Owning a dog can be expensive, with costs ranging from food and medical expenses to emergency care and property damage. Therefore, it is not surprising that there is a strong correlation between median income and dog ownership. On average, workers in Manhattan have a median yearly income of $107,962, significantly higher than workers in other boroughs. The Bronx has the lowest median income of all the boroughs, which may explain why it has significantly fewer registered dogs than its population would suggest.

Data Science...

Our project aims to predict the number of dogs in each zip code based on various socio-economic factors. The response variable is the dog count, while the predicting variables are based on the data overview contents in each notebook

To create our model, we followed these steps:
To create our model, we started with a linear regression baseline model and added lasso regularization to remove trivial features. Using cross-validation, we determined that a degree of 1 was optimal for polynomial regression. Grid search was used to find the best max_depth for a decision tree regressor and plot feature importance. We then used a bagging regressor and gradient boosting, optimizing the latter with grid search. Our best model nested an optimal gradient boost regressor within an Ada Boost regressor, achieving a test accuracy of 0.769.