| # 🎓 Student Dropout Prediction Dataset — EDA Assignment |
|
|
| > **By Tomer Bash** | Data Science Course — Assignment #1 |
|
|
| --- |
|
|
| ## 📹 Presentation Video |
|
|
| Presentation Video link - https://youtu.be/KyafBx9W7Qg |
|
|
| <video src="presentation.mp4" controls="controls" style="max-width: 720px;"></video> |
|
|
| --- |
|
|
| ## 📌 Dataset Overview |
|
|
| | Property | Details | |
| |----------|---------| |
| | **Source** | Kaggle | |
| | **Rows** | 4,424 students | |
| | **Features** | 35 columns | |
| | **Target Variable** | `Target` — `Graduate`, `Enrolled`, `Dropout` | |
| | **Task Type** | Multi-class Classification | |
|
|
| The dataset contains demographic, financial, academic, and application-related information about students at a Portuguese higher education institution. The primary goal is to understand and predict **student dropout before first-year academic results are available**. |
|
|
| --- |
|
|
| ## 🎯 Research Goal |
|
|
| > **Can we predict whether a student will drop out — early, before first-year grades are available?** |
|
|
| --- |
|
|
| ## 🧹 Part 1: Data Cleaning & Preparation |
|
|
| - **No missing values** — `df.isnull().sum()` returned 0 for all columns. |
| - **No duplicate rows** — `df.duplicated().sum()` returned 0. |
| - All 35 features are numeric (int/float) — verified with `df.info()`. |
| - Target column contains exactly 3 classes: `Graduate`, `Enrolled`, `Dropout`. |
| - **Class imbalance noted:** |
| - Graduate: 2,209 |
| - Dropout: 1,421 |
| - Enrolled: 794 |
| - **Outlier detection:** Age at enrollment shows right-skewed distribution with outliers at high ages — identified using a boxplot. |
| - **Grade anomaly detected:** Significant spike at grade = 0 in 1st semester grades — investigated and documented below. |
|
|
| --- |
|
|
| ## 📊 Part 2: Exploratory Data Analysis |
|
|
| ### 📈 Target Distribution |
|
|
| The dataset is imbalanced. Graduate is the majority class, followed by Dropout, and Enrolled is the smallest group. This imbalance must be handled carefully in any downstream modeling. |
|
|
| --- |
|
|
| ### 👤 Age at Enrollment |
|
|
| - Distribution is **right-skewed** — most students enroll young, but a long tail of older students exists. |
| - Boxplot reveals several **outliers at high ages** (up to ~70). |
| - **The Dropout group has a higher average age** than the Graduate group — age at enrollment is a risk factor. |
|
|
| | Group | Avg Age at Enrollment | |
| |---|---| |
| | Graduate | ~22 | |
| | Enrolled | ~23 | |
| | Dropout | ~26 | |
|
|
| --- |
|
|
| ### 📚 1st Semester Grades — The 0-Grade Anomaly |
|
|
| - Grades follow a near-normal distribution overall. |
| - A **massive spike at grade = 0** was detected — far more students received exactly 0 than expected. |
| - Upon investigation: some students with grade = 0 **still graduated**. |
| - **Conclusion:** The 0-grade entries likely represent administrative records (late withdrawals, course deregistrations), not pure academic failure. This is a data quality issue to flag for modeling. |
| - The anomaly is **concentrated in Course 2 (Animation degree)**, though the proportion is consistent with class sizes. |
|
|
| --- |
|
|
| ### 💰 Financial Factors — Primary Predictor |
|
|
| Financial health is the strongest non-academic predictor of student outcomes. |
|
|
| | Financial Factor | Effect on Dropout | |
| |---|---| |
| | **Tuition fees not up to date** | Overwhelmingly **high** dropout rate | |
| | **Scholarship holder** | Drastically **reduces** dropout risk | |
| | **University debtor** | Significantly **increases** dropout risk | |
|
|
| > Students not keeping up with tuition fees show the single highest dropout signal in the entire dataset. |
|
|
| --- |
|
|
| ### ⚧ Gender |
|
|
| - **Female students (0)** show a slightly higher graduation rate and lower dropout rate than male students (1). |
| - The difference is modest but consistent across the dataset. |
|
|
| --- |
|
|
| ### 🏛️ Application Mode |
|
|
| - Top 5 most common application modes were analyzed. |
| - **Application Mode 12 (Over 23 years old)** has the highest dropout rate — consistent with the age finding above. |
| - Standard routes (1st phase general contingent) perform best overall. |
|
|
| --- |
|
|
| ### 🔥 Correlation Heatmap |
|
|
| Features analyzed: Age at enrollment, Tuition fees up to date, Scholarship holder, 1st & 2nd semester grades, GDP, Target (encoded 0=Dropout, 1=Enrolled, 2=Graduate). |
|
|
| **Key findings:** |
| - ✅ **Grades** and **financial stability** are the strongest positive predictors of graduation. |
| - ❌ **Age at enrollment** correlates negatively with graduation. |
| - 📉 **GDP** shows almost no direct linear impact — individual finances matter more than the macro economy. |
|
|
| --- |
|
|
| ## 🏁 Conclusion |
|
|
| ### 1. 💰 Financial Stability — Primary Predictor |
| Scholarships drastically reduce dropout. Unpaid tuition and university debt dramatically raise it. Crucially, these signals exist **before** first-year grades are available — enabling early identification of at-risk students. |
|
|
| ### 2. 📚 Grades & the 0-Grade Anomaly |
| High grades are a strong positive predictor of graduation. The spike at grade = 0 is likely administrative, not academic — some 0-grade students still graduated. |
|
|
| ### 3. 🎂 Older Students Are at Higher Risk |
| Age at enrollment correlates negatively with graduation. Students admitted via Mode 12 ("Over 23") have the highest dropout rates — likely due to work and family pressures. |
|
|
| ### 4. 🌍 Macro Economy Has No Direct Linear Effect |
| National GDP shows near-zero correlation with the target. Individual financial circumstances matter far more. |
|
|
| --- |
|
|
| ## 📁 Repository Contents |
|
|
| | File | Description | |
| |------|-------------| |
| | `dataset.csv` | The full student dataset (4,424 × 35) | |
| | `Assignment_1_EDA_Tomer_Bash.ipynb` | Full EDA notebook with code & visualizations | |
| | `README.md` | This file | |
| | `presentation.mp4` | Video walkthrough (2–3 min) | |
|
|
| --- |
|
|
| ## 🛠️ Technologies Used |
|
|
| `Python 3` · `pandas` · `numpy` · `matplotlib` · `seaborn` · `Jupyter Notebook` |
|
|
| --- |
|
|
| *Dataset originally sourced from Kaggle. Uploaded to HuggingFace as part of Data Science Course Assignment #1.* |
|
|
