EDA Capstone Loyihasi

🎯 Maqsad

1-oyning yakunlovchi loyihasi. Real datasetda **to'liq Exploratory Data Analysis (EDA)**bajarib, professional darajadagi reporttayyorlaysiz. Bu sizning portfolio'ngizdagi birinchi ish bo'ladi.

Loyiha brief

Dataset tanlovi (bittasini tanlang)

**Tavsiya:**Birinchi marta — House Pricesyoki Titanic. Bular yaxshi hujjatlangan va Kaggle'da minglab kernel'lar bor.

EDA report'ning standart strukturasi

1. Project Overview (1 sahifa)

• Maqsad: nima uchun bu ma'lumotlarni tahlil qilamiz?
• Business question: javob izlanayotgan asosiy savol
• Dataset haqida qisqacha (manba, hajmi, ustunlar soni)

2. Data Loading va Initial Inspection

df = pd.read_csv("data.csv")
print(df.shape)          # nechta qator/ustun
print(df.dtypes)         # ustun tiplari
print(df.head())         # birinchi qatorlar
print(df.info())         # memory va null'lar
print(df.describe())     # statistik xulosa

3. Data Quality Check

• Missing values — har ustunda nechta NaN, % bo'yicha
• Duplicates — df.duplicated().sum()
• Data type issues — masalan, date string ko'rinishida
• Outliers — IQR yoki z-score usulida
• Value distributions — har bir categorical ustunda unique qiymatlar

# Missing values vizualizatsiyasi
import missingno as msno
msno.matrix(df)
msno.bar(df)

4. Univariate Analysis

Har bir ustunni alohida o'rganish:

• Numerical: histogram, KDE, box plot
• Categorical: count plot, value_counts
• Date: time series plot

5. Bivariate Analysis

• Ikki ustun orasidagi munosabat
• Num vs Num: scatter, correlation
• Cat vs Num: box plot, violin plot
• Cat vs Cat: cross-tabulation, stacked bar

6. Multivariate Analysis

• 3+ ustun aralashgan
• Pair plot(Seaborn)
• Heatmap(correlation matrix)
• Faceted plots(FacetGrid)

7. Target Variable Deep Dive

Agar supervised ML maqsadingiz bo'lsa:

• Target distribution
• Class imbalance (classification)
• Feature vs target munosabati

8. Feature Engineering Ideas

EDA jarayonida quyidagilarni qayd qiling:

• Yangi feature'lar yaratish g'oyalari (masalan, age * income)
• Transformatsiya zarur ustunlar (log, sqrt)
• Encoding strategiyalari (categorical → numerical)

9. Key Insights (BIZNES TILIDA)

• 5-10 ta asosiy topilma
• Har biri bitta gap, ortidan vizualizatsiya
• Storytelling: "Mijozlar 70% ehtimol bilan oxirgi 3 oyda qo'ng'iroq qilmagan bo'lsa, ketadi"

10. Conclusion va Next Steps

• EDA xulosasi
• Modelga o'tish uchun tavsiyalar
• Datadagi cheklovlar va xavflar

Texnik talablar

Tools

• Jupyter Notebookyoki VS Code(.ipynb)
• Pandas — data manipulation
• NumPy — hisob-kitoblar
• Matplotlib + Seaborn — vizualizatsiya
• missingno — missing data vizual
• pandas-profilingyoki ydata-profiling(avtomatik EDA report)

pip install pandas numpy matplotlib seaborn missingno ydata-profiling

Code quality

• Notebook'ni mantiqiy section'larga bo'lish (Markdown headings bilan)
• Har bir kod bloki oldida tushuntirish
• Function'larga ajratish (def plot_distribution(col))
• Reproducibility: random_state=42 doim aniq

Notebook strukturasi (har bo'lim alohida cell)

# 1. IMPORTS
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from pathlib import Path

pd.set_option("display.max_columns", 100)
sns.set_theme(style="whitegrid")

# 2. LOAD DATA
DATA_PATH = Path("../data/house_prices.csv")
df = pd.read_csv(DATA_PATH)

# 3. OVERVIEW
print(f"Shape: {df.shape}")
print(f"Memory: {df.memory_usage(deep=True).sum() / 1e6:.1f} MB")
df.head()

Deliverable (topshiriladigan ish)

GitHub repo'da quyidagilar bo'lishi kerak:

eda-house-prices/
├── README.md                       # Loyiha tavsifi, qanday ishga tushirish
├── notebooks/
│   └── 01_eda.ipynb               # Asosiy EDA
├── data/
│   ├── raw/                       # Dastlabki CSV (gitignore bilan)
│   └── processed/                 # Tozalangan dataset
├── reports/
│   ├── insights.md                # 5-10 ta key insights (markdown)
│   ├── figures/                   # PNG/PDF chart'lar
│   └── eda_report.html            # ydata-profiling output
├── src/
│   └── plotting.py                # Reusable plot funksiyalari
└── requirements.txt

README.md shabloni

# House Prices EDA

## Maqsad
Ames Housing datasetni tahlil qilib, uy narxiga ta'sir qiluvchi asosiy omillarni aniqlash.

## Asosiy topilmalar
- OverallQual eng kuchli korrelatsiyaga ega (0.79)
- GrLivArea (yashash maydoni) ikkinchi muhim feature (0.71)
- ...

## Qanday ishga tushirish
\`\`\`bash
pip install -r requirements.txt
jupyter notebook notebooks/01_eda.ipynb
\`\`\`

## Texnologiyalar
- Python 3.11
- pandas, numpy, matplotlib, seaborn

Evaluation criteria

O'zingizni baholash uchun:

Maqsad: kamida har mezonda 2 ball.

Referenslar

• Kaggle EDA notebooks(eng yaxshilarni o'rganing):
• Comprehensive Data Exploration with Python
• Titanic EDA + ML
• "Effective Data Storytelling" — Brent Dykes
• ydata-profilingdocs — docs.profiling.ydata.ai

Bonus mashqlar (extra credit)

1. Streamlit dashboard: EDA natijalarini interaktiv dashboard'ga aylantiring
2. Automated EDA: ydata-profiling yoki Sweetviz ishlatib avtomatik report yarating va manual EDA bilan solishtiring
3. Geographic visualization(agar dataset'da lat/long bo'lsa): Folium yoki Plotly bilan map yarating

✅ Loyihani topshirishdan oldin

• Notebook xatosiz to'liq run bo'ladi
• Har bir chart title, xlabel, ylabel, legendga ega
• Markdown bilan har bo'lim tushuntirilgan
• README aniq va to'liq
• GitHub'ga commit qilingan (notebook ham, charts ham)
• LinkedIn'ga post yozasiz (bu — sizning birinchi ML loyihangiz!)

Tabriklayman — birinchi katta qadam tugadi. Mashqlar bo'limidagi qo'shimcha praktikani ham bajaring.

So'ngra: Oy 2 — Klassik ML ga o'ting.