Two assignments for MATH F432 Applied Statistical Methods (BITS Pilani, Dubai Campus), covering interval estimation, hypothesis testing, multiple linear regression, one-way ANOVA, and a short classification extension — each with a reproducible Python script, a full Word/PDF report, and a summary Excel dashboard.
- Assignment 1 — Interval Estimation & Hypothesis Testing
- Assignment 2 — Regression, ANOVA & Classification
- Dashboard
- Repository Structure
- Installation
- Reproducing the Results
- Data Sources
- License
Dataset: healthcare-dataset-stroke-data.csv (Kaggle stroke prediction dataset).
Task: for stroke patients (n = 209 with valid BMI), (a) compute a 95% confidence interval for true mean BMI, and (b) test H0: μ = 28.5 kg/m² (the national average) against Ha: μ ≠ 28.5 at α = 0.05.
Result: sample mean BMI = 30.47, t(208) = 4.503 > critical value 1.971 → reject H0. 95% CI = (29.61, 31.33), which excludes 28.5, confirming the decision.
Since the raw BMI values are right-skewed (Shapiro–Wilk p < 0.0001) and fail a formal normality check, the analysis also includes a robustness extension: Cohen's d (effect size = 0.311, small-to-medium), a Wilcoxon signed-rank test (non-parametric, p = 3.6 × 10⁻⁴), and a 10,000-resample bootstrap CI (29.63, 31.33) — all three agree with the t-test, so the conclusion holds regardless of the normality violation.
Files: Asm1/Aditya_ASM_Assignment_1.py, Asm1/AdityaSharma_Assignment1_ASM.docx / .pdf.
Task: fit Sales ~ TV + Radio + Newspaper (200 observations), test overall and per-predictor significance, and check regression assumptions.
Result: the model is highly significant overall (F(3,196) = 605.4, p ≈ 8 × 10⁻⁹⁹, R² = 0.903). TV and Radio are significant positive predictors (p < 0.001 each); Newspaper is not significant (p = 0.954, its 95% CI (-0.011, 0.012) contains 0). VIFs are all below 3 (no multicollinearity), and a Breusch–Pagan test confirms homoscedasticity; residuals show only mild non-normality, not a concern at n = 200.
Task: test whether mean Petal Length differs across the 3 iris species, check ANOVA assumptions, and (extension) classify species from all four measurements.
Result: Petal Length differs significantly across species (F(2,147) = 1179.03, p ≈ 3 × 10⁻⁹¹, η² = 0.941 — a very large effect). Levene's test shows unequal variances across groups, so the analysis also reports Welch's ANOVA and the non-parametric Kruskal-Wallis test as robustness checks — both agree with the classical ANOVA. Tukey's HSD confirms every pairwise species comparison is significant. A Linear Discriminant Analysis (LDA) classifier trained on all four measurements reaches 97.8% test accuracy (44/45, stratified 70/30 split), misclassifying only one borderline versicolor/virginica flower.
Files: Asm2/advertising_regression.py, Asm2/iris_anova_classification.py, Asm2/AdityaSharma_Assignment2_ASM.docx / .pdf, Asm2/build_report.js (regenerates the Word report from the two scripts' results and figures).
ASM_Assignments_Dashboard.xlsx is a single Excel workbook summarizing all three analyses: a Dashboard sheet with KPI cards and four native charts (BMI CI, regression coefficients, ANOVA group means, LDA per-species F1-score), plus one detail sheet per analysis with the full numeric results. Regenerate it with python build_dashboard.py after re-running the three analysis scripts.
docs/index.html is a single self-contained HTML page (Chart.js via CDN, no backend, no build step) covering all three analyses: KPI cards, four charts, and full results tables (including the Tukey HSD post-hoc comparisons). It's generated by build_web_dashboard.py from the same source data as the Excel dashboard:
python Asm1/Aditya_ASM_Assignment_1.py
python Asm2/advertising_regression.py
python Asm2/iris_anova_classification.py
python build_dashboard.py
python build_web_dashboard.pyOpen docs/index.html directly in a browser to preview it locally. To publish it, push this repo to GitHub and enable Settings → Pages with source main branch, /docs folder — no separate account or deploy step needed. The same folder also deploys as-is to Vercel (npx vercel deploy docs --prod) or Netlify (drag docs/ onto app.netlify.com/drop) if you'd rather use those.
.
├── README.md
├── LICENSE
├── requirements.txt
├── build_dashboard.py # builds ASM_Assignments_Dashboard.xlsx
├── build_web_dashboard.py # builds docs/index.html (GitHub Pages / Vercel / Netlify)
├── ASM_Assignments_Dashboard.xlsx
│
├── docs/
│ └── index.html # live, self-contained web dashboard (Chart.js)
│
├── Asm1/
│ ├── Aditya_ASM_Assignment_1.py # BMI one-sample t-test + robustness checks
│ ├── healthcare-dataset-stroke-data.csv
│ ├── fig.png # generated by the script
│ ├── AdityaSharma_Assignment1_ASM.docx
│ └── AdityaSharma_Assignment1_ASM.pdf
│
└── Asm2/
├── advertising_regression.py # Part A: multiple linear regression
├── iris_anova_classification.py # Part B: one-way ANOVA + LDA
├── build_report.js # builds the combined Word report (Node + docx)
├── advertising.csv
├── iris.csv
├── iris/ # original UCI iris.data / iris.names
├── fig_*.png # generated by the two scripts
├── AdityaSharma_Assignment2_ASM.docx
└── AdityaSharma_Assignment2_ASM.pdf
All Python scripts resolve their input/output paths relative to their own location (BASE_DIR = os.path.dirname(os.path.abspath(__file__))), so the repo runs correctly from any clone location.
Not tracked in git (see .gitignore): Asm2/SPSSStatistics.exe (905MB SPSS installer — over GitHub's 100MB file limit and a licensed installer that shouldn't be redistributed) and the raw iris.zip archives (duplicates of Asm2/iris/, which is what's actually used).
git clone https://github.com/<your-username>/asm-assignments.git
cd asm-assignments
python -m venv venv
source venv/bin/activate
pip install -r requirements.txtThe Word-report builder for Assignment 2 (Asm2/build_report.js) additionally needs Node.js and the docx npm package:
cd Asm2
npm install docx
node build_report.js# Assignment 1
python Asm1/Aditya_ASM_Assignment_1.py
# Assignment 2
python Asm2/advertising_regression.py
python Asm2/iris_anova_classification.py
# Dashboards (after the three scripts above)
python build_dashboard.py
python build_web_dashboard.pyEach script prints its full statistical output to the console and saves its figures as .png files next to the script.
- Stroke prediction dataset:
healthcare-dataset-stroke-data.csv, originally published on Kaggle. Used here for Assignment 1's BMI hypothesis test. - Advertising dataset:
advertising.csv, the classic TV/Radio/Newspaper spend vs. Sales dataset from An Introduction to Statistical Learning (James, Witten, Hastie & Tibshirani). - Iris dataset: R.A. Fisher (1936), via the UCI Machine Learning Repository (
Asm2/iris/iris.data,iris.names).
The code and written reports in this repository are released under the MIT License (see LICENSE). The three datasets above each carry their own separate terms from their original sources and are not covered by this license.