Campus Placement Prediction - Logistic Regression, Decision Tree, K-Nearest Neighbors (KNN)

This project aims to predict a student's placement status based on academic and extracurricular factors such as CGPA, internship experience, and project involvement. Three machine learning classifiers were built and evaluated: Logistic Regression, Decision Tree, and K-Nearest Neighbors (KNN). Each model was tuned and assessed for accuracy to determine which best predicts placement outcomes.

Dataset

The dataset used in this analysis can be found here.

Dataset Attributes

• Academic Performance: CGPA, SSC Marks, HSC Marks
• Skills Ratings: Aptitude Test Score, Soft Skills Rating
• Extracurricular Factors: Internship Experience, Project Involvement, Placement Training
• Placement Status: Target variable indicating whether a student is placed or not

Exploratory Data Analysis (EDA)

Exploratory Data Analysis was conducted using Python libraries such as pandas, seaborn, and matplotlib. Key insights include:

• Placement Analysis: Out of 10,000 students, 40% are placed while 60% are not, showing a significant proportion of unplaced students.

• Box Plot Analysis: Box plots for marks-related attributes (CGPA, Aptitude Test Score, Soft Skills Rating, SSC Marks, HSC Marks) indicate no significant outliers, suggesting a well-distributed and reliable dataset for prediction.

• Scatter Plot Analysis: Scatter plots reveal that high CGPA, SSC, and HSC marks alone do not guarantee placement, implying that placement outcomes are influenced by factors beyond academic scores.

• Chi-Square Hypothesis Testing: The Chi-Square test results indicate a statistically significant difference in placement rates between students who received placement training and those who did not.

Project Steps

1. Data Preprocessing

• Load Dataset: Imported and explored the dataset to understand its structure.
• Handle Missing Values: Checked and handled any missing data.
• Encode Categorical Features: Categorical data was converted into numerical format using appropriate encoding.
• Feature Selection: Selected relevant features for modeling.
• Data Splitting: Split the dataset into training (80%) and testing sets (20%).
• Feature Scaling: Standardized numerical features to optimize model performance.

2. Model Building and Evaluation

Logistic Regression

• Parameter Tuning: Tested different values of max_iter (100, 200, 300, 400, 500).

• Best Accuracy Score: 79.8% for all max_iter values, indicating stable performance.

  

Decision Tree

• Parameter Tuning: Tested different values of max_depth (1-10).

• Best Accuracy Score: 77.6% with max_depth = 3.

  

K-Nearest Neighbors (KNN)

• Parameter Tuning: Tested different values of k (5-20).

• Best Accuracy Score: 79.1% at k = 15, which balanced accuracy and stability.

  

3. Model Comparison and Analysis

Model	Best Parameter	Accuracy Score
Logistic Regression	max_iter = 100	79.8%
Decision Tree	max_depth = 3	77.6%
KNN	k = 15	79.1%

4. Evaluation Summary

• Logistic Regression: Best-performing model with an accuracy of 79.8%. It offers stability and interpretability, although it may miss non-linear relationships.
• Decision Tree: Scored 77.6% and shows signs of overfitting with increasing depth beyond max_depth = 4.
• KNN: Competitive with an accuracy of 79.1%, though computationally intensive and sensitive to k-values. Recommended k = 15 for the best balance between accuracy and stability.

5. Conclusion

Logistic Regression proved to be the most effective model, achieving the highest accuracy and offering a consistent performance across different iterations. KNN also showed strong performance, though it required careful tuning of k. Decision Tree, despite capturing non-linear relationships, performed the least effectively and exhibited risks of overfitting.

Dependencies

• Python
• Pandas
• Seaborn
• Matplotlib
• SciPy
• Scikit-Learn