Project Overview
This project was completed during my 6-month part-time position as a Working Student at Siemens Turkey. I developed a time series forecasting solution to predict future sales of electrical products, aiming to support the company’s demand planning and inventory optimisation through data-driven insights.
Dataset
- Duration: 5 years of historical monthly sales data
- Scope: Electrical product categories across different business lines
- Structure: Univariate time series (monthly aggregated)
Methodology
- Model: SARIMA (Seasonal ARIMA)
- Tools: Python (pmdarima, statsmodels, pandas, matplotlib)
- Steps:
- Trend and seasonality analysis using decomposition
- Grid search for optimal SARIMA (p,d,q)(P,D,Q)s parameters
- Residual diagnostics to ensure model assumptions
- 3-month sales forecasting
Results
- The SARIMA model successfully captured yearly seasonality and sales volatility.
- Forecasts were well aligned with real business cycles and peaks in demand.
- Visualisations and confidence intervals were provided for planning and reporting.