SuperStore Sales Analysis and Prediction: A Data Science Approach

Project Overview

This project showcases how data science techniques can be applied to analyze SuperStore’s sales data, uncover actionable insights, and predict future performance. Leveraging Python and its robust libraries, I explore key drivers of sales and profitability, develop predictive models, and provide recommendations to optimize revenue and profitability.

---

Objective

The primary goal is to analyze sales data and build machine learning models to:

1. Predict sales revenue and profit margins.
2. Classify orders as profitable or non-profitable.
3. Provide data-driven strategies for improving business performance.

---

Problem Statement

SuperStore’s sales data holds valuable insights that can drive strategic decisions. This project addresses the following:

1. Identifying Financial Drivers: Analyze customer segments, discounting practices, product categories, and regional performance to determine which factors influence revenue and profitability the most.
2. Predictive Modeling: Build machine learning models (linear regression and logistic regression) to forecast sales and classify profitability.
3. Actionable Recommendations: Provide practical suggestions for discount strategies, customer targeting, and inventory management.

---

Dataset Overview

The dataset includes records of transactions with the following key features:

Most Relevant Features

• Order Date (order_date): Captures seasonality and sales trends.
• Ship Date (ship_date): Indicates logistics efficiency and potential impact on customer satisfaction.
• Product Name (product_name): Highlights products with high sales or profitability.
• Segment (segment): Differentiates sales across customer segments.
• Category (category) and Subcategory (subcategory): Group products for broader insights into sales patterns.
• Region (region), City (city), State (state): Show geographical variations in sales performance.
• Discount (discount): Directly affects sales volume and profitability.
• Quantity (quantity): Strongly correlated with sales revenue.
• Profit (profit): Key metric for profitability analysis.

Less Relevant Features

• Order ID (order_id): Unique identifier; not predictive of sales.
• Zip Code (zip): Too granular; higher-level geographical features are more useful.

---

Key Project Features

1. Sales Prediction: Linear regression is used to forecast sales revenue.
2. Profitability Analysis: Logistic regression is employed to classify orders as profitable or non-profitable.
3. Actionable Insights: Trends and patterns are extracted to guide strategic decisions.
4. Recommendations: Data-driven strategies for optimizing discounts, targeting customers, and managing inventory.

---

Current Progress

Logistic Regression Model

• Objective: Classify orders as profitable (1) or non-profitable (0).
• Current Accuracy:
  Using validation data, the logistic regression model achieves an accuracy of 97.25%:

  print((y_val == profit_decision).mean())  # 0.9724862431215607
  

Cross-Validation

• Implemented: K-Fold Cross-Validation with 10 folds for more robust evaluation.
• AUC Scores:
• Mean AUC across folds: 0.997 ± 0.001.

Fine-Tuning

• Regularization Strength (C): Tuned using cross-validation:
• Optimal C Value: 0.5 with AUC of 0.999 ± 0.000.

Findings

• Key Drivers:
• Discount: Significant impact on profitability and sales volume.
• Region and Customer Segment: Clear differences in sales performance based on geography and segment.
• Logistic Regression: Highly effective for classifying profitable orders with near-perfect AUC scores.
• Regularization: Fine-tuning C improved model performance.