Data Science Techniques for Exploring Data

• Average employee age is around 37 years. It has a high range, from 18 years to 60, indicating good age diversity in the organization.
• At least 50% of the employees live within a 7 km radius from the organization. However there are some extreme values, seeing as the maximum value is 29 km.
• The average monthly income of an employee is USD 6500. It has a high range of values from 1K-20K, which is to be expected for any organization's income distribution. There is a big difference between the 3rd quartile value (around USD 8400) and the maximum value (nearly USD 20000), showing that the company's highest earners have a disproportionately large income in comparison to the rest of the employees. Again, this is fairly common in most organizations.
• Average salary hike of an employee is around 15%. At least 50% of employees got a salary hike 14% or less, with the maximum salary hike being 25%.
• Average number of years an employee is associated with the company is 7.
• On average, the number of years since an employee got a promotion is 2.18. The majority of employees have been promoted since the last year.

• The age distribution is close to a normal distribution with the majority of employees between the ages of 25 and 50.
• The percentage salary hike is skewed to the right, which means employees are mostly getting lower percentage salary increases.
• MonthlyIncome and TotalWorkingYears are skewed to the right, indicating that the majority of workers are in entry / mid-level positions in the organization.
• DistanceFromHome also has a right skewed distribution, meaning most employees live close to work but there are a few that live further away.
• On average, an employee has worked at 2.5 companies. Most employees have worked at only 1 company.
• The YearsAtCompany variable distribution shows a good proportion of workers with 10+ years, indicating a significant number of loyal employees at the organization.
• The YearsInCurrentRole distribution has three peaks at 0, 2, and 7. There are a few employees that have even stayed in the same role for 15 years and more.
• The YearsSinceLastPromotion variable distribution indicates that some employees have not received a promotion in 10-15 years and are still working in the organization. These employees are assumed to be high work-experience employees in upper-management roles, such as co-founders, C-suite employees and the like.
• The distributions of DailyRate, HourlyRate and MonthlyRate appear to be uniform and do not provide much information. It could be that daily rate refers to the income earned per extra day worked while hourly rate could refer to the same concept applied for extra hours worked per day. Since these rates tend to be broadly similiar for multiple employees in the same department, that explains the uniform distribution they show.

•  Total work experience, monthly income, years at company and years with current manager are highly correlated with each other and with employee age which is easy to understand as these variables show an increase with age for most employees.
• Years at company and years in current role are correlated with years since last promotion which means that the company is not giving promotions at the right time.

The feature importance plot for the base model and tuned model are quite similar

• the model seems to suggest that OverTime, MonthlyIncome, Age, TotalWorkingYears and DailyRate are the most important features.
• Other important features are DistanceFromHome, StockOptionLevel, YearsAt Company and NumCompaniesWorked.

 Note: Blue leaves represent the eligible class i.e. y[1] and the orange leaves represent the non-eligible class i.e. y[0]. Also, the more the number of observations in a leaf, the darker its color gets.

• Employees who are doing overtime with low salary and low age have a chance of leaving the company, as they might feel overworked and underpaid and might be looking for better opportunities.
• Employees who are doing overtime with low salary and are not research scientists have a high chance of attriting.
• Employees, even if they have an income over 3751.5 units but working as sales executives and and living far from home have a high chance of attriting.
• Another segment of people are who are doing overtime, with ages younger than 33.5 and not working as junior research scientists, have a greater chance of attrition. This implies that the model suggests except for the junior research scientist role, everyone who is young has a high tendency to attrite.
• Employees who have over 2.5 years of work experience but low work life balance and low percentage hike also tend to attrite, probably as they are seeking a more balanced life.
• Employees who are not doing overtime, have low experience and are working as junior research scientists have a small chance of attriting. These employees are comfortable or loyal to the organization.
• Numcompaniesworked also seems to be an important variable in predicting if an employee's likely to attrite.