Data Model Modifications and Updates#
The ML model from phase 1 was completely scrapped for a variety of reasons, including a lack of impact the output would have on the user. The new model uses exponential regression (reasoning discussed below) to output the predicted number of immigrants expected to apply for asylum to a given country in a given year. This was used on a different dataset from the same API.
Below is the list of tables that we have currently and whether or not this data will be sourced or generated via mockaroo:
Appointments#
Source: Generated via Mockaroo
Description: Contains appointment details for users and volunteers.
AttendeeEvents#
Source: Generated via Mockaroo
Description: Records of events attended by users and volunteers.
CommunityEvents#
Source: Generated via Mockaroo
Description: Information about various community events organized for or involving users and volunteers.
Countries#
Source: Sourced
Description: List of countries and their details.
CountryStats#
Source: Sourced
Description: Statistical data related to countries, including population and asylum application statistics.
Dependents#
Source: Generated via Mockaroo
Description: Information about dependents of users
## Posts#
Source: Generated via Mockaroo
Description: User-generated posts and related information.
RefugeeCohort#
Source: Sourced
Description: Details and statistics about refugee cohorts.
Users#
Source: Generated via Mockaroo
Description: User profiles including personal information and relevant attributes.
Volunteers#
Source: Generated via Mockaroo
Description: Profiles of volunteers.
Dataset#
This is the dataset used to train the exponential regression model, taken from the same API used in phase 2. This dataset, however, gives information regarding the age and gender demographics of asylum seekers from 2010-2022 as opposed to the varying types of immigration applications used in phase 2. The advantage to this change in dataset was that the user archetypes discussed in phase 2 would likely have greater affinity to the predictive output generated to this dataset. For example, Tanya Bracker could use the information about the demographics surrounding asylum seekers to better allocate funds for schools (just one of many real world examples). Because these data are from a public API, they are sourced, and not generated.
This graph depicts the residuals of the exponential regression. There is clearly heteroscedasticity, exemplified by the lack of residuals randomly scattered around zero, an indicator of homoscedasticity. A future improvement for this model which will be investigated shortly is a box-cox transformation. This was ultimately hypothesized as a possible solution for the relatively poor model outcomes after viewing a Q-Q plot to investigate the distribution of residuals.
This plot is detailing that the residuals appear to be non-normally distributed. Because this is an indicator of poor model fitting, a box-cox transformation will be performed as an attempt to normalize the distribution of residuals.Features used for ML and reasoning#
Country of Asylum (coa): Demographic patterns can differ significantly between countries. Including this variable enables the model to give more accurate information which is subjective to a specific country.
Year : Demographic patterns can change over time due to factors like economic conditions, political instability, conflicts, and policy changes. The model should capture these temporal changes to make accurate future predictions.
Age: Age is a critical demographic factor that influences health care needs, educational requirements, and economic participation. Breaking down the population into age groups allows the model to capture differences in demographic profiles and needs across various age brackets.
Gender: Gender is a fundamental demographic factor that can affect migration patterns, health needs, and social vulnerabilities. Including gender allows the model to differentiate demographic patterns between males and females.
API Routes and Explanations#
City Council API Routes#
GET /city_council Purpose: Retrieve all community events from the database.
POST /council_add_event Purpose: Create a new community event.
PUT /city_council/communityEvent Purpose: Update an existing community event.
DELETE /city_council/communityEvent/
GET /city_council Purpose: Retrieve all appointments from the database.
GET /city_council/
DELETE /city_council/delete_bulletin/<post_id> Purpose: Delete a specific bulletin post based on its post ID.
GET /city_council/bulletin Purpose: Retrieve all bulletin posts.
Migrant API Routes#
GET /migrant/appointments/
GET /migrant/show_appt/
POST /migrant Purpose: Create a new appointment for a migrant.
PUT /migrant/appointment Purpose: Update the appointment details for a specific migrant.
DELETE /migrant/appointment_delete/
GET /migrant/posts Purpose: Retrieve all posts from the database.
GET /migrant/
POST /migrant/add_post Purpose: Create a new post for a migrant.
PUT /migrant/post Purpose: Update a specific post for a migrant.
DELETE /migrant/post/
GET /migrant/events Purpose: Retrieve all community events from the database.
Immigration Official API Routes#
GET /immigration_official/population Purpose: Retrieve the population from the database based on countryID.
GET /immigration_official/countryStats Purpose: Retrieve all country statistics from the database.
POST /immigration_official/newStat Purpose: Create a new country statistic entry.
PUT /immigration_official/editCountryStat Purpose: Update an existing country statistic.
DELETE /immigration_official/deleteCountryStat Purpose: Delete a country statistic entry based on countryID.
GET /immigration_official/numApplications Purpose: Retrieve the number of applications for a specific country in a given year.
GET /immigration_official/applicationTrends Purpose: Retrieve the overall trend of applications for a specific country in a given year.
Demonstration of Working Model#
Surprises/setbacks#
The main difficulty when designing this model was that the linear regression model (to put it bluntly) was terrible. The r2 value was 0.02, and the mse was over 10,000,000. The idea of trying exponential regression came due to the intuition of Dr. Gerber after observing the appearance of an exponential slope in a comparable graph conducted as a proof of concept in the phase 2 blog post:
