3 of 65

Executive Summary

To analyze the data provided to build a Machine Learning based solution that can help in shortlisting the candidates having higher chances of VISA approval - with the help of a classification model, the follow steps were followed;

Sanity checks on the dataset/Data overview.
Exploratory data analysis/EDA was done.
Data preprocessing was done.
Model building and performance checks.
Model assumptions check.
Calculate performance metrics and create confusion matrix for different models.
Compare all models.
Feature importance of XGBoost Hyperparameter Tuned Model.

After building the final model to get the decision tree results, the insights and recommendation are as follows;

For the model, we have;

Education of employee - employee with a doctorate degree have 65% chance of getting visa certified, and employee with high school certification has over 65% of getting visa denied.
Unit of wage - employee with non-hourly pay has 70% chance of getting visa certified, and employee with hourly pay has 65% of getting visa denied.
Continent - employee with work experience and from Europe has 75% and 80% chance of getting visa certified compared to employee with no work experience have 50% chance of getting visa denied.
Region of employment - employees from Midwest and South have 70% chances of getting visa certified.
Attributes such as; full time or part time position, require job training, prevailing wage and year of establishment do not have much impact for visas to get certified or denied.
We built a model that can capture over 80% of the information while making predictions, the findings can help build a suitable profile of candidates to facilitate the process of visa.

4 of 65

Business Problem Overview and Solution Approach

Problem Statement

Business communities in the United States are facing high demand for human resources, but one of the constant challenges is identifying and attracting the right talent, which is perhaps the most important element in remaining competitive. Companies in the United States look for hard-working, talented, and qualified individuals both locally as well as abroad.

The Immigration and Nationality Act (INA) of the US permits foreign workers to come to the United States to work on either a temporary or permanent basis. The act also protects US workers against adverse impacts on their wages or working conditions by ensuring US employers' compliance with statutory requirements when they hire foreign workers to fill workforce shortages. The immigration programs are administered by the Office of Foreign Labor Certification (OFLC).

OFLC processes job certification applications for employers seeking to bring foreign workers into the United States and grants certifications in those cases where employers can demonstrate that there are not sufficient US workers available to perform the work at wages that meet or exceed the wage paid for the occupation in the area of intended employment.

Solution approach/methodology

In FY 2016, the OFLC processed 775,979 employer applications for 1,699,957 positions for temporary and permanent labor certifications. This was a nine percent increase in the overall number of processed applications from the previous year. The process of reviewing every case is becoming a tedious task as the number of applicants is increasing every year.

The increasing number of applicants every year calls for a Machine Learning based solution that can help in shortlisting the candidates having higher chances of VISA approval. The objective is to analyze the data provided and, with the help of a classification model:
Facilitate the process of visa approvals.
Recommend a suitable profile for the applicants for whom the visa should be certified or denied based on the drivers that significantly influence the case status.

5 of 65

EDA Results

Key results from EDA
Majority of the employees come from the the continent Asia.
Majority of the employee have Bachelor’s degree.
About 58% have job experience and 42% have no job experience.
More foreign workers intend applying for jobs in the Northeast region (28%) followed by South (28%) and West (26%) regions.
90% Unit prevailing wage falls under Yearly, 9% falls under Hourly.
Employee with High School are most denied and employee with Doctorate degrees are most certified or offered visa.
Please mention answers to the insight-based questions provided
Education of employee - employee with a doctorate degree have 65% chance of getting visa certified, and employee with high school certification has over 65% of getting visa denied.
Unit of wage - employee with non-hourly pay has 70% chance of getting visa certified, and employee with hourly pay has 65% of getting visa denied.
Continent - employee with work experience and from Europe has 75% and 80% chance of getting visa certified compared to employee with no work experience have 50% chance of getting visa denied.
Region of employment - employees from Midwest and South have 70% chances of getting visa certified.

We built a model that can capture over 80% of the information while making predictions, the findings can help build a suitable profile of candidates to facilitate the process of visa.

Link to Appendix slide on data background check

6 of 65

EDA Results

View the top 5 rows of the dataset View the last 5 rows of the dataset

Link to Appendix slide on data background check

7 of 65

EDA Results

Checking the shape/dimension of the dataset.

The dataset has 25480 rows and

12 columns.

Link to Appendix slide on data background check

8 of 65

EDA Results

Checking the data types of the columns for the dataset.

There are 5 columns of the dtype object,

1 column of the dtype float64,

and 2 columns of the dtype int64.

Link to Appendix slide on data background check

9 of 65

EDA Results

Checking for duplicate values.

Link to Appendix slide on data background check

10 of 65

EDA Results

Statistical summary of the data.

Link to Appendix slide on data background check

11 of 65

EDA Results

Checking for negative values in the

employee column.

Link to Appendix slide on data background check

12 of 65

EDA Results

Checking the count of each unique

category in each of the the

categorical variables.

Link to Appendix slide on data background check

13 of 65

EDA Results

Checking unique values in the

mentioned column.

Link to Appendix slide on data background check

14 of 65

EDA Results

Checking unique values in the

mentioned column.

Link to Appendix slide on data background check

15 of 65

EDA Results

Univariate Analysis

Observations on number of employees

- The distribution of the number of employees

is right-skewed.

- The boxplot shows that there are lots of

outliers to the right.

Link to Appendix slide on data background check

16 of 65

EDA Results

Univariate Analysis

Observations on prevailing wage

Visual analysis of both distributions shows

- right-skewed.

- the mean is around USD 70,000.

- outliers in the income bracket between

USD 200,000 to USD 300,000

Link to Appendix slide on data background check

17 of 65

EDA Results

Rows with less than 100 prevailing wage

Link to Appendix slide on data background check

18 of 65

EDA Results

Count of the values in the mentioned column

Link to Appendix slide on data background check

19 of 65

EDA Results

Univariate Analysis

Observations on continent

Visual analysis of bar plot shows

- Majority of the employees come from the

the continent Asia.

Link to Appendix slide on data background check

20 of 65

EDA Results

Univariate Analysis

Observations on education of employee

Visual analysis of bar plot shows

- Majority of the employee have Bachelor’s degree.

Link to Appendix slide on data background check

21 of 65

EDA Results

Univariate Analysis

Observations on job experience

Visual analysis of the bar plot shows

- About 58% have job experience and 42% have

No job experience.

Link to Appendix slide on data background check

22 of 65

EDA Results

Univariate Analysis

Observations on region of

employment

Visual analysis of the bar plot shows

- More foreign workers intend applying

For jobs in the Northeast region (28%)

followed by South (28%) and West (26%)

regions.

Link to Appendix slide on data background check

23 of 65

EDA Results

Univariate Analysis

Observations on unit wage

Visual analysis of the bar plot shows

- 90% Unit prevailing wage falls under Yearly,

9% falls under Hourly.

Link to Appendix slide on data background check

24 of 65

EDA Results

Univariate Analysis

Observations on case status

Visual analysis of the bar plot shows

- About 67% of cases are certified or approved and

33% are denied.

Link to Appendix slide on data background check

25 of 65

EDA Results

Bivariate Analysis

Observations on

correlation heat map

Visual analysis of the heatmap shows

- There is positive correlation between

no_of_employees, yr_of_estab and

Prevailing_wage.

Link to Appendix slide on data background check

26 of 65

EDA Results

Bivariate Analysis

Observations on education impact

on visa certification

Visual analysis of the barplot shows

- employee with High School are most

denied and employee with Doctorate degrees

are most certified or offered visa.

Link to Appendix slide on data background check

27 of 65

EDA Results

Bivariate Analysis

Observations on

educational background requirement

for different regions

Visual analysis of the heatmap shows

- The requirement for Bachelors in more

in the South, requirement for Doctorate

is more in the West, requirement for

High School is more in the South and

Requirement for Master’s is more in the

Northeast region.

Link to Appendix slide on data background check

28 of 65

EDA Results

Bivariate Analysis

Observations on percentage of

visa certifications across each region

Visual analysis of the stacked barplot shows

- Midwest region have the highest visa

certifications, Island and Northeast

have the lowest.

Link to Appendix slide on data background check

29 of 65

EDA Results

Bivariate Analysis

Observations on how visa status

vary across different continents.

Visual analysis of the barplot shows

- Europe and Africa leads getting their

visa certified compared to other

continents.

Link to Appendix slide on data background check

30 of 65

EDA Results

Bivariate Analysis

Observations on job experience and

case status

Visual analysis of the barplot shows

- applicants who have job experience are

more likely to get visa certified. About 80% of

those who have job experience got certified.

Link to Appendix slide on data background check

31 of 65

EDA Results

Bivariate Analysis

Observations on job experience and

require job training

Visual analysis of the barplot shows

- applicants who have job experience requires

less job training.

Link to Appendix slide on data background check

32 of 65

EDA Results

Bivariate Analysis

Observations on distribution of

prevailing wage and case status

Visual analysis of the distribution plot shows

- the median prevailing wage for certified

applications is slightly higher compared

to the denied applications.

Link to Appendix slide on data background check

33 of 65

EDA Results

Bivariate Analysis

Observations on region of employment

and prevailing wage

Visual analysis of the distribution plot shows

- Midwest and Island have slightly higher

prevailing wages compared to rest.

Link to Appendix slide on data background check

34 of 65

EDA Results

Bivariate Analysis

Observations on unit wage and

case status

Visual analysis of the barplot shows

- applicant with yearly unit wage have

high chance of getting certified followed

by applicants with weekly and monthly

unit wage. Hourly unit wage applicants

have the lowest chance of getting certified.

Link to Appendix slide on data background check

35 of 65

Data Preprocessing

Check for outliers in the data

Visual analysis of the boxplot shows

- few outliers are present in the data.

36 of 65

Model Performance Summary

Checking model performance on training set

Visual analysis of the confusion matrix for train data

shows zero errors 0% on training set and overfitting.

Link to Appendix slide on model assumptions

37 of 65

Model Performance Summary

Checking model performance on test set

Visual analysis of the confusion matrix for test data

shows overfitting. Needs improvement.

Link to Appendix slide on model assumptions

38 of 65

Model Performance Summary

Hyperparameter Tuning - Decision Tree

Visual analysis of the confusion matrix for training data

on tuned estimator.

Link to Appendix slide on model assumptions

39 of 65

Model Performance Summary

Checking model performance on test set

Visual analysis of the confusion matrix for test data

on tuned estimator shows reduction in overfitting,

F1 score for both training and test set is 0.80.

Link to Appendix slide on model assumptions

40 of 65

Model Performance Summary

Bagging - Model Building and Hyperparameter

Visual analysis of the confusion matrix for training data

shows overfitting.

Link to Appendix slide on model assumptions

41 of 65

Model Performance Summary

Checking model performance on test set

Visual analysis of the confusion matrix for test data

on bagging classifier shows reduction in overfitting,

F1 score for both training and test set is 0.80

Link to Appendix slide on model assumptions

42 of 65

Model Performance Summary

Hyperparameter Tuning - Bagging Classifier

Visual analysis of the confusion matrix for training data

shows overfitting.

Link to Appendix slide on model assumptions

43 of 65

Model Performance Summary

Checking model performance on test set

Visual analysis of the confusion matrix for test data

on bagging classifier shows big difference between

training and test data.

Link to Appendix slide on model assumptions

44 of 65

Model Performance Summary

Random Forest

Visual analysis of the confusion matrix for training data

shows overfitting.

Link to Appendix slide on model assumptions

45 of 65

Model Performance Summary

Random Forest

Visual analysis of the confusion matrix for test data

on bagging classifier showing F1 as 0.80

Link to Appendix slide on model assumptions

46 of 65

Model Performance Summary

Random Forest

Visual analysis of the confusion matrix for test data

on bagging classifier showing F1 as 0.80

Link to Appendix slide on model assumptions

47 of 65

Model Performance Summary

Hyperparameter Tuning - Random Forest

Visual analysis of the confusion matrix for training data

shows it has generalized.

Link to Appendix slide on model assumptions

48 of 65

Model Performance Summary

Hyperparameter Tuning - Random Forest

Visual analysis of the confusion matrix for test data

shows good precision. F1 scores are 0.84 and 0.82

on training and test data. No overfitting.

Link to Appendix slide on model assumptions

49 of 65

Model Performance Summary

Hyperparameter Tuning - AdaBoost Classifier

Visual analysis of the confusion matrix for training data

shows it has generalized.

Link to Appendix slide on model assumptions

50 of 65

Model Performance Summary

Hyperparameter Tuning - AdaBoost Classifier

Visual analysis of the confusion matrix for test data

shows good precision. F1 scores are 0.78 and 0.78

on training and test data. Model has good precision and high recall.

Link to Appendix slide on model assumptions

51 of 65

Model Performance Summary

Gradient Boosting Classifier

Visual analysis of the confusion matrix for training data

shows it has generalized.

Link to Appendix slide on model assumptions

52 of 65

Model Performance Summary

Gradient Boosting Classifier

Visual analysis of the confusion matrix for test data

shows model has generalized performance, with F1 score

0.83 and 0.82 for training and test set.

Link to Appendix slide on model assumptions

53 of 65

Model Performance Summary

Hyperparameter Tuning - Gradient Boosting Classifier

Visual analysis of the confusion matrix for train data

Link to Appendix slide on model assumptions

54 of 65

Model Performance Summary

Hyperparameter Tuning - Gradient Boosting Classifier

Visual analysis of the confusion matrix for test data shows

no much difference after hyperparameter tuning.

Link to Appendix slide on model assumptions

55 of 65

Model Performance Summary

XGBoost Classifier

Visual analysis of the confusion matrix for train data

Link to Appendix slide on model assumptions

56 of 65

Model Performance Summary

XGBoost Classifier

Visual analysis of the confusion matrix for test data shows

generalized performance, but will further tune the

hyperparameters to see if any further improvement.

Link to Appendix slide on model assumptions

57 of 65

Model Performance Summary

Hyperparameter Tuning - XGBoost Classifier

Visual analysis of the confusion matrix for train data

Link to Appendix slide on model assumptions

58 of 65

Model Performance Summary

Hyperparameter Tuning - XGBoost Classifier

Visual analysis of the confusion matrix for test data shows

generalized performance, with F1 score of 0.82 for both

training and test set.

Link to Appendix slide on model assumptions

59 of 65

Model Performance Summary

Stacking Classifier

Visual analysis of the confusion matrix for train data

Link to Appendix slide on model assumptions

60 of 65

Model Performance Summary

Hyperparameter Tuning - XGBoost Classifier

Visual analysis of the confusion matrix for test data shows

generalized performance, with F1 score of 0.82 for both

training and test set, as we have in XGBoost model (no difference)

Link to Appendix slide on model assumptions

61 of 65

Model Performance Summary

Model Performance Comparison and Final Model Selection

Training performance comparison

Link to Appendix slide on model assumptions

62 of 65

Model Performance Summary

Model Performance Comparison and Final Model Selection

Test performance comparison

Findings shows Decision Tree, Bagging Classifier (Default and Tuned), Random Forest (Default and Tuned) were found to overfit the training set.
Decision Tree (Tuned), Random Forest (Tuned), AdaBoost Default), Gradient Boost (Default and Tuned), XGBoost (Default and Tuned) and Stacking (Default) gave a generalized performance on both training and testing datasets.
Stacking Classifier has the highest F1 score.

Link to Appendix slide on model assumptions

63 of 65

Model Performance Summary

Important features of the final model

The findings shows education of employee is

the most important attribute having an influence

on visa certifications.

Other important attributes are; employee having a

prior job experience, prevailing wage and continent

of the employee.

Link to Appendix slide on model assumptions

64 of 65

Actionable Insights and Recommendations

Based on the EDA and Stacking Classifier model, the following were observed as important factor for visas to get certified or get denied;

Education of employee - employee with a doctorate degree have 65% chance of getting visa certified, and employee with high school certification has over 65% of getting visa denied.
Unit of wage - employee with non-hourly pay has 70% chance of getting visa certified, and employee with hourly pay has 65% of getting visa denied.
Continent - employee with work experience and from Europe has 75% and 80% chance of getting visa certified compared to employee with no work experience have 50% chance of getting visa denied.
Region of employment - employees from Midwest and South have 70% chances of getting visa certified.

Attributes such as; full time or part time position, require job training, prevailing wage and year of establishment do not have much impact for visas to get certified or denied.

We built a model that can capture over 80% of the information while making predictions, the findings can help build a suitable profile of candidates to facilitate the process of visa.

1 of 65

2 of 65

3 of 65

4 of 65

5 of 65

6 of 65

7 of 65

8 of 65

9 of 65

10 of 65

11 of 65

12 of 65

13 of 65

14 of 65

15 of 65

16 of 65

17 of 65

18 of 65

19 of 65

20 of 65

21 of 65

22 of 65

23 of 65

24 of 65

25 of 65

26 of 65

27 of 65

28 of 65

29 of 65

30 of 65

31 of 65

32 of 65

33 of 65

34 of 65

35 of 65

36 of 65

37 of 65

38 of 65

39 of 65

40 of 65

41 of 65

42 of 65

43 of 65

44 of 65

45 of 65

46 of 65

47 of 65

48 of 65

49 of 65

50 of 65

51 of 65

52 of 65

53 of 65

54 of 65

55 of 65

56 of 65

57 of 65

58 of 65

59 of 65

60 of 65

61 of 65

62 of 65

63 of 65

64 of 65

65 of 65