Automotive Sales Forecasting System

📋 Project Overview

An automotive manufacturer required accurate sales forecasting across the vehicle lifecycle (production → wholesale → retail) to optimize production planning. We developed a mathematical modeling system using time series analysis and convolution-based prediction. The system models delay distributions between lifecycle stages and uses PySpark for distributed processing across multiple regions and vehicle series.

🚀 Key Features

• Multi-stage Lifecycle Modeling: Models complete vehicle flow from build → wholesale → retail with time delay distributions
• Convolution-based Forecasting: Transforms delay distributions into future sales predictions using kernel convolution
• Mathematical Optimization: Scipy L-BFGS-B algorithm for parameter optimization across multiple metrics

💻 Project Detail

1. Data Processing: PySpark-based aggregation of historical vehicle transaction data by region, series, and model year
2. Delay Distribution Modeling: Extract and model time gaps between production, wholesale, and retail stages
3. Prediction Pipeline: Convolution-based forecasting from build schedules to wholesale and retail sales
4. Parameter Optimization: L-BFGS-B optimization to minimize combined RMSE across sales stages

📊 Project Impact

High Model Interpretability & Optimization Capability:

• Mathematical model structure provides full transparency into prediction logic and parameters
• Enables what-if scenario analysis: adjust production parameters to simulate annual sales impact
• Business stakeholders can directly optimize production planning based on model outputs

Superior Forecasting Accuracy:

• Incorporating business-specific characteristics of each lifecycle stage (build delays, wholesale patterns, retail demand)
• Achieved significantly lower RMSE compared to standard time series models (ARIMA, Prophet, etc.)
• Multi-stage modeling captures domain knowledge that generic models cannot learn

🛠️ Technology Stack

Core Technologies:
  - PySpark (Distributed Data Processing)
  - Scipy (Mathematical Optimization)
  - NumPy (Convolution & Numerical Computing)
  - Pandas (Data Manipulation)

Modeling Approach:
  - Time Series Analysis
  - Convolution-based Forecasting
  - L-BFGS-B Optimization

This project demonstrates mathematical modeling and optimization techniques in automotive sales forecasting, providing interpretable and actionable predictions for production planning.