Student-Dropout-Prediction-University-of-Kigali

AUTHOR — NTAMBARA RUKAKA STEVEN

SUPERVISOR — DR MUSABE JEAN BOSCO

This notebook provides a comprehensive exploratory data analysis of the Student Dropout dataset, exploring factors that contribute to student dropout rates. Dataset Characteristics: 10,000 student records 19 features (12 numeric, 7 categorical) Binary target: Dropout (0=No, 1=Yes) 23.54% dropout rate

The dataset contains student data from University of Kigali Main Campus (and comparable institutional records) used to predict Dropout status, addressing the complex reasons behind student attrition. It includes 10,000 samples, focusing on scenarios that help predict student dropout risk. Attributes include age, gender, family income, academic performance, attendance, stress index, and behavioral factors.

Student Dropout Prediction — University of Kigali

Big Data Analytics for Early Dropout Risk Detection & Intervention

---

📌 Overview

University of Kigali (UoK) — like most higher-education institutions — loses a meaningful share of enrolled students to dropout every academic year. Today, that loss is only visible after the fact: a student stops appearing on attendance sheets, stops submitting assignments, and is eventually withdrawn — with no unified, real-time view of who is at risk and why, until it is too late to intervene.

This project delivers:

1. A structured Big Data analytics pipeline — cleaning, imputation, outlier treatment, and feature engineering — applied to a 10,000-record student dataset (student_dropout_dataset_v3.csv).
2. Exploratory data analysis (EDA) uncovering the demographic, academic, and behavioral patterns behind dropout at UoK-scale institutions.
3. A prediction layer (Logistic Regression, Decision Tree, Random Forest, SVM) that flags students at risk of dropout early enough for academic/student-affairs staff to intervene — before withdrawal is finalised.
4. A full scientific research manuscript reporting methodology, results, and evidence-based recommendations for an institutional early-warning system.

---

🎯 Problem / Causes of Dropout

Student dropout is rarely caused by one factor — it is the cumulative result of academic, financial, behavioral, and psychological pressures. Based on this dataset and the wider literature, the leading causes examined are:

• Declining academic performance — falling GPA, Semester GPA, and CGPA relative to a student's own historical trend.
• Elevated stress levels — the strongest non-academic correlate of dropout in this dataset (Stress Index).
• Poor attendance and low engagement — irregular class attendance and low weekly study hours.
• Financial pressure — low family income, and the need to balance a part-time job alongside studies.
• Logistical burden — long commute/travel time to campus.
• Delayed academic conduct — late assignment submissions as an early behavioral warning sign.
• Lack of institutional support signals — absence of scholarship support for financially vulnerable students.

Institutionally, the core problem is that:

• Academic records, attendance systems, and financial-aid data live in separate systems with no unified, cross-cutting view of at-risk students.
• Advisors have no early-warning signal for which students are drifting toward dropout, or why (grades? stress? finances? logistics?).
• Interventions are applied reactively and uniformly, rather than being targeted at the students who need them most.

✅ Solution

A data pipeline and analytics layer that:

1. Cleans and standardises institutional student records (demographic, academic, behavioral) into a single analysis-ready dataset.
2. Engineers derived indicators — GPA_Trend (short-term academic trajectory) and Engagement_Score (composite attendance + study-hours signal) — that sharpen prediction beyond raw fields.
3. Trains and compares multiple machine-learning classifiers to score each student's probability of dropout.
4. Surfaces the dominant, evidence-ranked risk factors (via feature importance) so advisors know what to act on, not just who to contact.
5. Documents the full pipeline, findings, and recommendations in a reproducible research manuscript suitable for institutional adoption.

---

🔗 University of Kigali Digital Properties

Relevant UoK digital properties this analytics work is designed to integrate with, for future live data collection (source: uok.ac.rw):

Property	URL	Role in this project
Main website	https://uok.ac.rw	Institutional context, public-facing reporting
Admissions	https://admissions.uok.ac.rw	Enrolment-funnel context for incoming-student risk baselining
Online Application	https://apply.uok.ac.rw	Application-stage demographic/financial-aid data source
"UoK Online Learning Portal"	https://learn.uok.ac.rw/my/	Source of attendance, submission, and engagement behavioural data
Student Portal (MIS)	https://mis.uok.ac.rw	Student records, GPA/CGPA, registration & academic-year status
Student Portal (MyCampus)	https://mycampus.uok.ac.rw	Student self-service; potential channel for risk-flag notifications
E-Learning Portal (alias)	https://elearning.uok.ac.rw	Secondary LMS access point
Digital / E-Library	http://ebooks.uok.ac.rw	Supplementary engagement signal (resource access)
Institutional Repository	https://repository.uok.ac.rw	Research & academic content engagement
Academic Calendar	https://uok.ac.rw/academic-calendar/	Context for term/semester start dates, aligning risk scoring to term timing

Social channels (for advisor-outreach/notification integrations): Facebook · X (Twitter) · YouTube · LinkedIn · Instagram

⚠️ URLs above were verified against the live UoK website at the time of writing and may change; always confirm against https://uok.ac.rw before wiring up live data feeds.

---

🏗️ Data Source Architecture

Student Records (MIS)          LMS / Moodle Logs             Financial-Aid / Admissions
   (GPA, CGPA, Semester,          (Attendance, login             (Family_Income,
    Department, Age,               frequency, assignment          Scholarship status)
    Gender, Parental_Ed)           delay days)
        │                              │                                │
        └──────────────┬───────────────┴────────────────┬───────────────┘
                        ▼                                ▼
              Unified Student Analytics Dataset (student_dropout_dataset_v3.csv)
                        │
                        ▼
              ┌─────────────────────────┐
              │  Preprocessing Pipeline  │
              │  • Missing-value imputation
              │  • IQR outlier capping
              │  • Feature engineering
              │    (GPA_Trend, Engagement_Score)
              └───────────┬─────────────┘
                          ▼
              ┌─────────────────────────┐
              │  Exploratory Analysis    │
              │  (12 charts: distribution,
              │   correlation, by-department,
              │   by-semester, stress, etc.)
              └───────────┬─────────────┘
                          ▼
              ┌─────────────────────────┐
              │  Machine Learning Models │
              │  Logistic Regression ·   │
              │  Decision Tree ·          │
              │  Random Forest · SVM      │
              └───────────┬─────────────┘
                          ▼
              ┌─────────────────────────┐
              │  Dropout Risk Score /    │
              │  Advisor Early-Warning   │
              │  Dashboard (future work) │
              └─────────────────────────┘

At UoK scale (10K records), this pipeline runs on a single machine with pandas/scikit-learn. For institution-wide, multi-year, multi-campus data (Kacyiru, Remera, Musanze), the identical logic can be re-implemented on Apache Spark (PySpark + MLlib) with data staged in HDFS, near-real-time student events streamed via Apache Kafka, and risk scores served through SQL data marts or MongoDB collections to an advising dashboard.

---

⚙️ Setup and Configuration

1. Clone & set up the environment

git clone https://github.com/your-org/Student-Dropout-Prediction-UOK.git
cd Student-Dropout-Prediction-UOK
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt

2. Configure environment variables (.env)

DATASET_PATH=data/student_dropout_dataset_v3.csv
OUTPUT_DIR=outputs/
MIS_API_BASE_URL=https://mis.uok.ac.rw
MIS_API_TOKEN=your_mis_api_token
LMS_BASE_URL=https://learn.uok.ac.rw
LMS_API_TOKEN=your_moodle_webservice_token
RANDOM_SEED=42

3. Run the analytics pipeline

python pipeline.py

This performs data cleaning → imputation → outlier capping → feature engineering → EDA chart generation → model training/evaluation, and writes all outputs to outputs/ (cleaned dataset, figures/, results_report.json).

4. Generate the research manuscript

node build_paper.js

Produces Student_Dropout_Research_Paper.docx from the latest results_report.json and figures/.

---

🗂️ Dataset — student_dropout_dataset_v3.csv

Records: 10,000 students · Attributes: 19 · Overall dropout rate: 23.5%

Column	Description
Student_ID	Unique student identifier
Age, Gender, Parental_Education	Demographics
Family_Income	Annual household income
Internet_Access	Home internet access (Yes/No)
Study_Hours_per_Day	Average daily self-study hours
Attendance_Rate	Percentage class attendance
Assignment_Delay_Days	Average late-submission days
Travel_Time_Minutes	Commute time to campus
Part_Time_Job	Holds part-time employment (Yes/No)
Scholarship	Receives scholarship (Yes/No)
Stress_Index	Self-reported stress score
GPA, Semester_GPA, CGPA	Academic performance measures
Semester	Academic year (Year 1–4)
Department	Academic department
Dropout	Target: 1 = dropped out, 0 = retained

Engineered features added during preprocessing:

Feature	Description
GPA_Trend	Semester_GPA − CGPA — short-term academic trajectory
Engagement_Score	Weighted composite of normalised attendance rate and study hours

See outputs/cleaned_dataset.csv for the cleaned, imputed, outlier-capped, feature-engineered dataset used for modeling, and outputs/results_report.json for the full machine-readable results summary.

---

🖼️ Screenshots & Visual References

University of Kigali — Kacyiru (main), Remera, and Musanze campuses

Analytics outputs generated by this project (in outputs/figures/):

Figure	Description
fig1_class_distribution.png	Class distribution of the Dropout target
fig2_dropout_by_department.png	Dropout rate by academic department
fig3_dropout_by_semester.png	Dropout rate by academic year
fig4_cgpa_distribution.png	CGPA distribution: retained vs. dropped out
fig5_attendance_boxplot.png	Attendance rate vs. dropout status
fig6_correlation_heatmap.png	Correlation matrix of numeric features
fig7_stress_violin.png	Stress Index distribution by dropout status
fig8_job_scholarship.png	Dropout rate by part-time job & scholarship status
fig9_model_comparison.png	Accuracy / precision / recall / F1 / ROC-AUC across all 4 models
fig10_roc_curves.png	ROC curves for all classifiers
fig11_confusion_matrix.png	Confusion matrix — Random Forest
fig12_feature_importance.png	Random Forest feature-importance ranking

---

🏗️ Project Structure

Student-Dropout-Prediction-UOK/
├── data/
│   └── student_dropout_dataset_v3.csv     # Raw dataset (input)
├── outputs/
│   ├── cleaned_dataset.csv                # Cleaned/engineered dataset (output)
│   ├── results_report.json                # Machine-readable summary of all findings
│   ├── figures/                           # 12 generated EDA + model charts (PNG)
│   └── Student_Dropout_Research_Paper.docx # Full scientific manuscript (final deliverable)
├── app/
│   ├── analytics/
│   │   ├── mis_client.py                  # MIS (student records) API client (future work)
│   │   └── moodle_client.py               # Moodle Web Services API client (future work)
│   ├── models/
│   │   └── dropout_predictor.py           # Dropout classification model wrapper
│   └── routes/
│       └── dashboard.py                   # Staff-facing risk dashboard (future work)
├── pipeline.py                            # Full preprocessing + EDA + modeling script
├── build_paper.js                         # Research-manuscript generation script (docx)
├── tests/
├── requirements.txt
├── .env.example
└── README.md

---

▶️ How to Run

git clone https://github.com/your-org/Student-Dropout-Prediction-UOK.git
cd Student-Dropout-Prediction-UOK
cp .env.example .env                 # fill in MIS/LMS credentials if using live data
pip install -r requirements.txt
python pipeline.py                   # cleans data, runs EDA, trains models, writes outputs/
node build_paper.js                  # builds the full research-paper docx

Output	Location
Cleaned dataset	outputs/cleaned_dataset.csv
Charts	outputs/figures/*.png
Results summary (JSON)	outputs/results_report.json
Research manuscript	outputs/Student_Dropout_Research_Paper.docx
Live LMS (reference)	https://learn.uok.ac.rw/my/
University site (reference)	https://uok.ac.rw

pytest tests/ -v --cov=app

---

🧰 Environment & Tool Versions

Tool	Version	Purpose
Python	3.11+	Runtime
pandas	3.0+	Data wrangling & cleaning
numpy	2.4+	Numeric computation
scikit-learn	1.8+	Machine learning (Logistic Regression, Decision Tree, Random Forest, SVM)
matplotlib	3.10+	Visualization
seaborn	0.13+	Statistical visualization
Node.js + docx (npm)	20+	Research-manuscript (.docx) generation
Moodle Web Services API	REST/JSON	LMS behavioural data source (learn.uok.ac.rw), future live integration
PostgreSQL	16	Analytics warehouse (future live deployment)
Apache Spark / Hadoop / Kafka	—	Scale-up path for institution-wide, multi-year data (see Data Source Architecture)
Docker	25+	Containerisation (future live deployment)

---

📈 Key Findings Snapshot

• Overall dropout rate: 23.5% (2,354 of 10,000 students).
• Strongest predictors: CGPA, Semester GPA, GPA (all negatively correlated with dropout, r ≈ −0.44 to −0.46).
• Strongest non-academic predictor: Stress Index (r ≈ +0.25).
• Logistic Regression achieved the best F1-score (0.585) and ROC-AUC (0.821); Random Forest achieved the highest accuracy (78.0%).
• Demographic factors (Gender, Internet Access, Parental Education, Family Income) showed minimal influence relative to academic and well-being indicators.

Full methodology, tables, and discussion are documented in outputs/Student_Dropout_Research_Paper.docx.

---

📚 Reference Links

• University of Kigali — https://uok.ac.rw
• UoK Online Learning Portal (LMS) — https://learn.uok.ac.rw/my/
• Admissions — https://admissions.uok.ac.rw
• Apply Online — https://apply.uok.ac.rw
• Student Portal (MIS) — https://mis.uok.ac.rw
• Student Portal (MyCampus) — https://mycampus.uok.ac.rw
• Digital/E-Library — http://ebooks.uok.ac.rw
• Institutional Repository — https://repository.uok.ac.rw
• Academic Calendar — https://uok.ac.rw/academic-calendar/
• scikit-learn documentation — https://scikit-learn.org/stable/documentation.html
• pandas documentation — https://pandas.pydata.org/docs/
• Apache Spark documentation — https://spark.apache.org/docs/latest/
• Moodle Web Services documentation — https://docs.moodle.org/dev/Web_services
• Ntambara Rukaka Steven Github — https://github.com/Ramjan0487/Student-Dropout-Prediction

---

📄 License

MIT — see LICENSE.

---

Not an official University of Kigali product. Built for academic/research purposes (Big Data Analytics coursework) using publicly available UoK URLs and assets as of July 2026.