Retail Sales Time Series Analysis

A complete time series analysis pipeline built with Python - from messy data to business insights.

Portfolio project demonstrating data cleaning, forecasting, analysis, and business storytelling for data analyst/scientist positions.

---

What This Project Does

Takes messy retail sales data spanning 3 years and transforms it into:

• Demonstrates data science workflow
• Shows data cleaning and validation skills
• Compares multiple forecasting approaches
• Extracts actionable business insights
• Creates visualizations

---

Run This

# Install dependencies
pip install -r requirements.txt

# Generate sample data
python generate_data.py

# Run complete analysis (3-5 minutes)
python run_all.py

---

Results from This Pipeline

Dataset Analyzed

• 1,095 daily sales records (2021-2023)
• $1.38 million total revenue
• 69,927 customers
• 3 store locations, 4 product categories

Data Quality Improvements

• 22 missing values → Interpolated using time-series methods
• 3 duplicate records → Removed
• 5 outliers → Capped using IQR method
• Inconsistent store names → Standardized

Forecasting Performance

Model	MAE ($)	RMSE ($)	MAPE (%)
Moving Average	187.31	214.69	14.28
ARIMA	196.18	223.36	14.97

Moving Average performed best for this dataset (lower error = better)

Business Insights Discovered

Growth:

• 14.8% YoY growth (2022)
• 14.4% YoY growth (2023)
• Consistent upward trend

Store Performance:

• Store_A: $537,027 (38.8% of sales)
• Store_B: $453,228 (32.8%)
• Store_C: $393,636 (28.4%)

Product Categories:

• Electronics: 30.7% of total sales
• Clothing: 26.5%
• Home & Garden: 23.8%
• Food & Beverage: 19.0%

Temporal Patterns:

• Best day: Sunday ($1,349 average)
• Best month: April (highest average sales)
• Weekend sales significantly higher than weekdays

---

Project Structure

time-series/
├── scripts/
│   ├── 01_data_cleaning.py
│   ├── 02_modeling_forecasting.py
│   ├── 03_analysis_insights.py
│   └── 04_executive_summary.py
│
├── data/                 # Generated outputs
│   ├── cleaned_sales_data.csv
│   ├── model_comparison.csv
│   ├── key_insights.txt
│   └── *.png            # 15+ visualizations
│
├── run_all.py
├── generate_data.py
└── requirements.txt

---

What Each Script Does

01_data_cleaning.py

Cleans messy retail data:

• Identifies and fixes missing values (2% of data)
• Removes duplicate dates
• Standardizes store names
• Handles outliers using statistical methods
• Validates data quality

Outputs: cleaned_sales_data.csv + 2 visualizations

---

02_modeling_forecasting.py

Builds forecasting models:

• Decomposes time series (trend, seasonality, residuals)
• Tests for stationarity (Augmented Dickey-Fuller)
• Trains Moving Average and ARIMA models
• Compares performance using MAE, RMSE, MAPE

Outputs: model_comparison.csv + 4 visualizations

---

03_analysis_insights.py

Extracts business insights:

• Calculates KPIs (revenue, transactions, customer metrics)
• Compares store performance across 3 locations
• Analyzes 4 product categories
• Identifies weekly and monthly patterns
• Performs statistical tests (t-tests for significance)

Outputs: key_insights.txt + 7 visualizations

---

04_executive_summary.py

Creates executive report:

• Business KPI dashboard
• Key findings summary
• Model performance visualization
• Strategic recommendations

Outputs: executive_summary_metrics.csv + 3 visualizations

---

Sample Visualizations

Executive Dashboard

Time Series Decomposition

Model Forecast Comparison

Store Performance

Category Distribution

Temporal Patterns

Monthly Sales Trend

Correlation Matrix

Data Cleaning: Before vs After

---

Skills Demonstrated

Technical Skills

• Python: pandas, numpy, matplotlib, seaborn, scipy
• Time Series: statsmodels, ARIMA, decomposition
• Statistics: hypothesis testing, correlation analysis, outlier detection
• Data Cleaning: handling missing values, duplicates, outliers
• Data Visualization: 15+ different chart types

Data Science Workflow

Raw Data → Clean → Explore → Model → Analyze → Report

1. Data quality assessment
2. Systematic cleaning and validation
3. Statistical analysis
4. Predictive modeling
5. Business insight extraction
6. Executive communication

To-do Future for this Project

• Interactive dashboard (Streamlit)
• Additional models (XGBoost, LSTM)
• A/B testing framework
• Anomaly detection
• REST API for predictions

---

Created: January 2026 Last Updated: January 2026