Data Preprocessing and EDA** and (Machine Learning & Deep Learning Models)** of Employee Attrition & Performance project:

Employee Attrition and Performance Prediction

📊 Overview

This project analyzes employee data to uncover patterns behind attrition and predict performance scores using machine learning (Random Forest & Linear Regression) and deep learning (Neural Network). It includes data preprocessing, exploratory analysis, statistical testing, classification, regression, and model evaluation.

📁 Dataset Description

File: employee_data.csv
Shape: 100 rows × 8 columns
Columns:

• EmployeeID: Unique identifier
• Name: Employee name
• Age: Employee age
• Department: Department name (e.g., HR, Sales)
• Salary: Annual salary (numeric)
• YearsAtCompany: Total years spent at the company
• PerformanceScore: Annual performance score (out of 100)
• Attrition: Whether the employee left the company (Yes/No)

✅ Step 1: Data Cleaning & Preprocessing

• Missing Values: Filled numeric columns using column mean
• Duplicates: Removed duplicate records
• Inconsistent Entries: Cleaned and standardized Department values
• Descriptive Stats: Summary generated using df.describe()

📊 Step 2: Exploratory Data Analysis (EDA)

Visualizations:

• Pairplot: To observe attrition patterns across features
• Correlation Heatmap: To assess feature relationships
• Boxplot: To detect outliers in Age, Salary, YearsAtCompany
• Attrition Probability by Department:
  • Sales: 35.89%
  • Engineering: 30.77%
  • HR: 23.08%
  • Marketing: 10.26%

Statistical Tests:

• Bayesian Inference:
  • P(Attrition | PerformanceScore) ≈ 0.395
• ANOVA (F-test):
  • Significant difference in performance scores across departments (p-value ≈ 2.56e-12)

🤖 Step 3: Machine Learning Models

Random Forest Classifier

• Target: Attrition
• Accuracy: 70%
• Precision/Recall (Class 1): 57%
• Confusion Matrix: Visualized using Seaborn heatmap

Linear Regression

• Target: PerformanceScore
• R² Score: 0.74
• Cross-Validated R² (5-fold): 0.73
• MAE: 0.36
• Visuals:
  • Predicted vs Actual
  • Residuals Plot

---

🧠 Step 4: Deep Learning - Neural Network (Keras)

• Architecture:
  • Dense(64, ReLU)
  • Dense(32, ReLU)
  • Dense(1, Linear)
• Loss: MSE
• Final Validation MAE: ~0.33
• Test MAE: ~0.49

---

📈 Trend Analysis

• Performance vs Tenure: Performance generally increases with YearsAtCompany
• Performance by Department & Attrition: Insights into department-specific attrition-performance patterns

---

🔧 Tools & Libraries Used

• Python Libraries: pandas, numpy, matplotlib, seaborn, scikit-learn, tensorflow, keras
• ML Models: Random Forest, Linear Regression
• DL Model: Keras Sequential Neural Network

---

📌 Future Improvements

• Hyperparameter tuning (Random Forest & Neural Net)
• Feature engineering with interaction terms
• Deployment using Streamlit or Flask
• SHAP/LIME for model explainability