-
STREAMLIT APP:
-
POWER BI DASHBOARD:
- Project Overview
- Live Demo
- Business Requirements
- Research Questions
- Dataset
- Methodology
- Key Findings
- Recommendations
- Ethical Approach
- Acknowledgments
This data analytics project investigates the factors that influence student academic performance, specifically examining which variables most strongly predict exam scores. Using a comprehensive dataset of student attributes including study habits, attendance patterns, and lifestyle factors, this study tests four interconnected hypotheses about academic success.
Explore the interactive dashboard yourself! Click the badge above or use this link: https://student-performance-dashboard-72e6c195af16.herokuapp.com/
Note: The app is hosted on a free Heroku dyno, so it may take 10-15 seconds to "wake up" if it hasn't been accessed recently. Please be patient!
BR1: Identify Key Performance Drivers
-
Requirement: Determine which student behaviors and characteristics have the strongest statistical relationship with exam performance.
-
Business Value: Schools and educational institutions can allocate resources (tutoring, counseling, intervention programs) to the factors that matter most.
-
Acceptance Criteria: Clear ranking of predictors with correlation coefficients and statistical significance levels.
BR2: Quantify Optimal Study-Sleep Balance
-
Requirement: Establish evidence-based thresholds for optimal study hours and sleep duration that maximize exam scores.
-
Business Value: Enable schools to provide specific, data-backed recommendations to students about healthy study habits.
-
Acceptance Criteria: Identification of "sweet spot" combinations with confidence intervals and group comparison results.
BR3: Develop Predictive Understanding
-
Requirement: Create a statistical model that can predict student exam performance based on behavioral and demographic factors.
-
Business Value: Early identification of at-risk students who may need additional support before exams.
-
Acceptance Criteria: Model with R² > 0.5 and identification of key predictive features.
BR4: Generate Actionable Student Profiles
-
Requirement: Segment students into meaningful groups based on study habits and performance patterns.
-
Business Value: Personalized recommendations and interventions for different student types (e.g., "high potential, low effort" vs. "high effort, low results").
-
Acceptance Criteria: Clear, interpretable student segments with distinct characteristics and performance outcomes.
BR5: Inform Educational Policy Recommendations
-
Requirement: Provide data-driven recommendations for school policies around homework load, scheduling, and wellness programs.
-
Business Value: Evidence-based decision making for administrators and education policymakers.
-
Acceptance Criteria: Concrete, actionable recommendations derived from statistical findings.
- Hours studied explains more variance in exam scores than attendance rate or sleep hours when analyzed separately. Analytical Approach = Simple linear Regression
- The combination of hours studied AND attendance rate predicts exam scores better than either variable alone. Analyticial Approach = Multiple regression with interaction.
- Students who study more than 20 hours per week AND get more than 7 hours of sleep score significantly higher than those meeting only one or neither threshold. Analytical Approach = ANOVA/ Group Comparison.
- Hours studied has the strongest correlation with exam scores among all continuous variables in the dataset. Analytical Approach = Correlation Matrix Analysis.
Source: Student Performance Factors (https://www.kaggle.com/datasets/lainguyn123/student-performance-factors) - Kaggle
Description: This dataset contains information on student demographics, study habits, parental involvement, and academic performance metrics.
Key Features:
- Hours Studied
- Attendance Rate
- Sleep Hours
- Previous Scores
- Extracurricular Activities
- Exam Score (Target Variable)
- Data Cleaning & Preprocessing:
- Handle missing values
- Check for outliers
- Convert data types
- Feature engineering (creating threshold groups for H3)
- Exploratory Data Analysis (EDA):
- Summary statistics
- Distribution visualizations
- Initial correlation exploration
- Hypothesis Testing:
- H1: Three simple linear regressions comparing R² values
- H2: Multiple regression with interaction term
- H3: Group comparisons using ANOVA/t-tests
- H4: Comprehensive correlation analysis with heatmap visualisation.
- Python (Recommended)
- Pandas (data manipulation)
- NumPy (numerical operations)
- Matplotlib/Seaborn (visualisations)
- SciPy/StatsModels (statistical testing)
- Scikit-learn (regression modeling)
This analysis will:
- Identify the strongest predictors of academic success
- Determine whether study hours alone matter more than combination factors
- Reveal if there's a "sweet spot" threshold for study and sleep
- Provide data-driven insights for students and educators
What We Found: Hours studied explained 19.8% (R² = 0.198) of the variance in exam scores, making it the only meaningful individual predictor. In contrast:
| Predictor | R² Value | Predictive Power |
|---|---|---|
| Hours Studied | 0.198 | Moderate |
| Attendance Rate | ~0.00 | None |
| Sleep Hours | 0.000 | None |
Why It Matters: When students ask "What's the most important thing I can do to improve my grades?" the data is clear: study time matters, while sleep hours and attendance alone (without study context) don't predict performance.
What We Found: While attendance alone doesn't predict scores, it plays a crucial moderating role. The interaction effect shows that:
- The relationship between study hours and exam scores changes depending on attendance levels
- Students with high attendance get more "bang for their buck" from each hour studied
Why It Matters: Showing up to class amplifies the effectiveness of studying. This means attendance policies aren't just about being present they directly impact how well study time translates to results.
What We Found: Students were grouped into four categories based on study hours and sleep:
| Group | Number of Students | Average Score | vs. "Low Both" |
|---|---|---|---|
| High Both (Study + Sleep) | 1,116 | 68.74 | +2.73 |
| High Study Only | 1,838 | 68.76 | +2.75 |
| High Sleep Only | 1,294 | 65.88 | +0.13 |
| Low Both | 2,130 | 65.75 | — |
Statistical Significance:
- ANOVA F-statistic = 308.39, p < 0.001
- Tukey HSD confirms: High Study Only and High Both groups score significantly higher than Low Both and High Sleep Only groups
Why It Matters: Study time drives success, not sleep alone. Students who study a lot score ~3 points higher regardless of whether they sleep a lot. However, the High Both group is the largest (high study + high sleep), suggesting successful students tend to maintain both habits.
What We Found: When comparing all factors, here's how they rank by correlation with exam scores:
| Rank | Factor | Correlation (r) | Strength |
|---|---|---|---|
| 1 | Attendance | 0.58 | Strong |
| 2 | Hours Studied | 0.45 | Moderate |
| 3 | Previous Scores | 0.16 | Weak |
| 4 | Tutoring Sessions | 0.03 | Very Weak |
| 5 | Physical Activity | 0.02 | Very Weak |
| 6 | Sleep Hours | -0.02 | None |
Why It Matters: This reveals something fascinating! While Hours Studied was the best individual predictor in regression (Finding 1), Attendance actually has a stronger correlation with exam scores (0.58 vs 0.45). This tells us:
- Attendance and study hours work together (as we saw in Finding 2)
- Students who attend class regularly also tend to study more
- Attendance might be a "gateway behavior" that enables effective studying
Figure 1: Correlation matrix showing relationships between all numerical variables
Figure 2: Exam scores across different study-sleep groups
Figure 3: Ranking of predictors by correlation strength with exam scores
| Finding | Key Insight | Business Requirement | Impact |
|---|---|---|---|
| 1 | Study hours explain 19.8% of score variance; sleep alone predicts nothing | BR1 | High |
| 2 | Attendance amplifies study effectiveness | BR3 | High |
| 3 | High study time = +2.7 points; High Both group is largest | BR2, BR4 | High |
| 4 | Attendance (r=0.58) and Study Hours (r=0.45) are top predictors | BR1, BR3 | High |
Based on these findings, we recommend:
| For | Recommendation | Evidence |
|---|---|---|
| Students | Prioritize 20+ weekly study hours AND attend class regularly | Findings 1, 2, 3 |
| Teachers | Track attendance closely—it identifies engaged students | Finding 4 |
| Schools | Communicate that attendance makes studying more effective | Finding 2 |
| Parents | Encourage consistent study routines over "cramming" | Finding 3 |
| Counselors | Target students with low attendance for early intervention | Finding 4 |
| Business Requirement | Status | Evidence |
|---|---|---|
| BR1: Identify Key Performance Drivers | YES | Attendance (r=0.58) and Study Hours (r=0.45) identified |
| BR2: Quantify Optimal Study-Sleep Balance | YES | High Both group (68.74) vs Low Both (65.75) |
| BR3: Develop Predictive Understanding | YES | Combined model shows attendance moderates study impact |
| BR4: Generate Actionable Student Profiles | YES | Four distinct groups with clear performance differences |
| BR5: Inform Educational Policy Recommendations | YES | 5 actionable recommendations provided |
- Statistical Analysis: Hypothesis testing, regression analysis, ANOVA, correlation matrices
- Python Programming: Pandas, NumPy, SciPy, StatsModels, Scikit-learn
- Data Visualization: Matplotlib, Seaborn, Streamlit dashboard development
- Critical Thinking: Translating business requirements into testable hypotheses
- Web Deployment: Heroku deployment, Git version control, README documentation
- Research Communication: Presenting complex findings to diverse audiences
When I started this project, I wanted to make sure I was handling student data responsibly. Here's how I thought about ethics throughout this analysis.
I used a dataset from Kaggle called "Student Performance Factors." It's publicly available and has a CC0 license, which means the person who shared it has waived all their rights. This gave me confidence that I wasn't stepping on anyone's toes by using it.
The first thing I checked was whether any individual student could be identified. I was relieved to find no names, no IDs, no addresses just general categories like "High" or "Medium" for family income, and scores without any student attached to them. So while the data tells us about patterns, it protects the privacy of real students.
I tried really hard not to overclaim. When I say things like "study hours predict exam scores," I know that's different from saying "study hours cause higher scores." There could be other factors at play like motivation or natural ability that I couldn't measure. My dashboards and README are careful to show correlations, not certainties.
This was important to me. I didn't want anyone reading this to think "if you don't study 20+ hours, you'll fail." That's not true, and that's not the message. Instead, I focused on patterns and possibilities—like how attendance makes studying more effective. The recommendations are about supporting students, not blaming them.
At its heart, this project is about understanding what helps students succeed. I wanted to create something that teachers, parents, and even students themselves could use to have better conversations about learning. Not to label anyone, but to find ways to help.
Dataset:
- Source: Student Performance Factors Dataset on Kaggle
- Creator: lainguyn123 (Kaggle user)
- Description: Comprehensive dataset containing information on student demographics, study habits, parental involvement, and academic performance metrics
- License: CC0: Public Domain
Inspiration & Learning:
- Streamlit documentation and community forums (dashboard development patterns)
- Kaggle community notebooks (EDA and statistical analysis techniques)
- Storytelling with Data (Cole Nussbaumer Knaflic) - dashboard layout and narrative principles
- Statology.org (statistical testing methodology references)
Tools:
- Dashboard deployed on Heroku (free tier)
- Interactive dashboard built with Streamlit (Python)
- Analysis completed using: Pandas, NumPy, Matplotlib, Seaborn, SciPy, StatsModels, Scikit-learn
- Development environment: Jupyter Notebook / VS Code
- Version control: Git / GitHub
- Documentation: Markdown
Technical Guidance & Support:
- Special thanks to an AI assistant from DeepSeek for real-time dashboard design consultation, statistical analysis guidance, and technical troubleshooting.
- This included: interactive filter implementation, Streamlit app optimization, README documentation refinement, and deployment support.
This project was developed as a portfolio piece to demonstrate statistical analysis, Python programming, and data visualization capabilities.
I extend my sincere gratitude to Code Institute for providing comprehensive data analytics education that formed the foundation of this project. The skills and methodologies learned throughout the course were instrumental in developing this data-driven analysis.
Special thanks to my tutors, Vasilica Pavaloi and Mark Briscoe, for their patient guidance and invaluable support in addressing numerous technical questions and challenges encountered during development.
I also wish to acknowledge my peers for their collaborative spirit, motivation, and shared insights that significantly contributed to the successful completion of this work.