Dashboard

Workforce overview & attrition signals
Live filtered view
— records
0 of 0 employees match

Headcount & Attrition by Department

Active vs. departed employees per department

Attrition Rate by Department

Share of employees who left, per department

Gender Distribution

Of records with gender recorded

Work Mode Split

Onsite / Hybrid / Remote

Overtime vs Attrition

Attrition rate for overtime vs no overtime

Age Distribution

Employee age bands

Satisfaction Score vs Attrition

Attrition rate at each satisfaction level (1–5)

Headcount by City

Employee count per office location

Years at Company

Tenure distribution

Employee Attrition Risk Prediction

Enter an employee profile to estimate attrition likelihood using the trained logistic regression model.
Fill in the employee profile and click
Predict Attrition Risk to see results.

Global Feature Importance

Absolute standardized logistic-regression coefficients — top drivers of predicted attrition

Confusion Matrix

Held-out test set ( employees)
Model Details

ROC Curve

True positive rate vs. false positive rate on the held-out test set — AUC =

About this dashboard

This dashboard is built entirely from the HRDBDA employee table extracted from the uploaded POWER_BI_2171.pbix data model — 5,514 employee records across 26 fields. One record with a missing Employee ID was removed, leaving employees analyzed. Everything you see — KPIs, charts, filters, the prediction model and its accuracy figures — is computed directly from this dataset. Nothing is simulated or invented.

Target variable & problem type:

Data quality notes

  • Salary is reported for only 674 of 5,513 employees (≈12%); the remaining values are 0 or negative and were treated as "not reported." Salary is therefore excluded from KPIs and the prediction model, and shown only as a coverage statistic.
  • City contained two spellings for the same city ("Banglore" / "Bengaluru") — these were merged.
  • Age, Performance Rating and Training Hours contained a small number of impossible values (e.g. age of 1 or 150, ratings outside 1–5, negative hours) — these were set to missing rather than guessed.
  • Remaining missing numeric values were filled with the column median; missing categorical values were filled with an explicit "Unknown" category rather than a guessed value.
  • 114 employees have no recorded Attrition value and were excluded from model training and from attrition-rate calculations.

Model & limitations

The prediction model is a logistic regression classifier trained in Python on the cleaned dataset (44 encoded features from 12 underlying fields), then exported as coefficients and run live in your browser — no server call is made when you click Predict.

Honest performance note: the model reaches ROC-AUC and accuracy on held-out data. That is meaningfully better than chance but far from perfect — this HR dataset has substantial missing/invalid data and attrition is only weakly explained by the fields available. Treat predictions as a directional risk signal, not a certainty.

Dataset snapshot

Fields used