01
Job satisfaction
Lower satisfaction appeared more frequently among employees who left the company.
Predictive Analytics
Logistic Regression
2026
Predicting
employee
attrition.
A predictive model designed to help Salifort Motors identify employee turnover risk and support more targeted retention decisions.
Employee turnover creates recruiting costs, productivity losses, and gaps in organizational knowledge.
Using employee workforce data, I conducted exploratory analysis and developed a logistic regression model to estimate which employees were more likely to leave.
01
Overview
Why do employees
leave?
Salifort Motors wanted to better understand which workplace factors were associated with employee turnover and whether departures could be predicted using historical workforce data.
The project focused on identifying meaningful attrition patterns, evaluating predictive performance, and translating model results into practical recommendations for HR and workforce planning.
Business Question
Can employee characteristics and workplace conditions be used to identify workers at greater risk of leaving?
02
Analytical Workflow
From employee data
to retention strategy.
The project followed the PACE framework, progressing from business planning and data preparation through exploratory analysis, model construction, evaluation, and stakeholder recommendations.
Each stage was designed to connect technical analysis with a clear business objective: reducing preventable employee turnover.
01
Plan
Defined the business problem, stakeholders, objectives, and measures of model success.
02
Prepare
Inspected, cleaned, standardized, and validated the employee dataset.
03
Explore
Analyzed employee characteristics, distributions, outliers, and attrition patterns.
04
Model
Built a logistic regression model to classify whether an employee would leave.
05
Recommend
Translated findings into retention strategies and future model improvements.
03
Employee Data
Ten variables.
One outcome.
The dataset contained 14,999 employee records across ten variables, with no missing values.
The target variable measured whether an employee remained with the company or left.
01
Satisfaction Level
Employee-reported job satisfaction
02
Performance Evaluation
Most recent evaluation score
03
Workload
Projects and average monthly hours
04
Company Tenure
Years spent at the organization
05
Employee Experience
Promotion and workplace-accident history
06
Role Context
Department and salary level
04
Exploratory Analysis
Turnover reflected
workplace conditions.
Exploratory analysis examined how satisfaction, workload, working hours, tenure, salary, department, promotion history, and performance differed between employees who stayed and employees who left.
The patterns suggested that employee turnover was associated with lower satisfaction and demanding workload conditions rather than one isolated factor.
02
Working hours
Longer working hours contributed to patterns associated with employee turnover.
03
Workload
Project demands and workload intensity were relevant when evaluating attrition risk.
05
Model Development
A transparent
baseline model.
Logistic regression was selected as an interpretable classification model suitable for estimating the probability that an employee would leave.
Numerical and categorical employee variables were prepared for modeling, then evaluated using accuracy, precision, recall, F1 score, ROC-AUC, and a confusion matrix.
01
→
Employee Data
02
→
Data Preparation
03
→
Logistic Regression
04
Model Evaluation
06
Model Results
Strong overall.
Limited recall.
The model produced solid overall classification performance, reaching 83.4% accuracy and a ROC-AUC of 0.83.
However, recall for employees who left was only 20.5%, meaning the model identified employees who stayed much more successfully than employees who departed.
83.4%
Overall logistic regression classification accuracy.
0.83
ROC-AUC, indicating good overall separation between employee outcomes.
20.5%
Recall for employees who left the company.
Accuracy alone did not tell the complete story. Because the model missed many actual departures, it would not be reliable enough to independently determine which employees require intervention.
The model is better positioned as an early screening tool that supports human review rather than replacing HR judgment.
07
Business Recommendations
Retention requires
human intervention.
The results point toward organizational conditions that may be addressed before an employee decides to leave.
Retention strategy should combine model-assisted screening with direct employee feedback and responsible management action.
01
Monitor workload
Identify teams and employees experiencing sustained high workloads or excessive monthly hours.
02
Improve work-life balance
Review staffing, project allocation, and management practices that may contribute to burnout.
03
Measure satisfaction
Use regular employee surveys and follow-up conversations to identify concerns before they become resignations.
08
Next Steps
Improving attrition
detection.
The next modeling priority should be increasing recall for employees who leave while maintaining acceptable precision and overall performance.
Potential improvements include addressing class imbalance, engineering additional features, tuning model parameters, and comparing logistic regression with decision trees, Random Forest, and gradient boosting models.
Ethical Use
Attrition predictions should support employee care—not become the sole basis for decisions about individual workers.
Project Files
Explore
Python Notebook
View the complete exploratory analysis, logistic regression workflow, and evaluation.Executive Summary
Stakeholder-facing overview of results, recommendations, and next steps.PACE Strategy
Project-planning framework documenting the analytical approach and decision process.Selected Work
Return to the complete portfolio project index.