In the dynamic landscape of grocery retail, accurate predictive sales are crucial for success. This project delves into a distinctive sales forecasting for Corporation Favorita in Ecuador, with the goal of securing a competitive edge and nurturing long-term growth.
- Title and brief project overview.
- Creation of the working environment and importation of necessary libraries.
- Collection and preparation of datasets from Database, OneDrive, and GitHub.
- Steps including data summary, handling missing values, data visualization, and distribution analysis.
- Selection, training, and evaluation of different models.
- Identification of best-performing models and recommendations.
- Collection of historical sales data, handling missing values, and ensuring data consistency.
- Use of interpolation technique to estimate missing data points, preserving trends and patterns.
-
Understanding Time Series Data:
- Analysis of total sales based on different groupings (City, Cluster, State, Store Type).
-
Visualization Techniques:
- Box plots, histograms, subplots, and bar plots to identify patterns and trends.
-
Stationarity:
- Application of Augmented Dickey-Fuller (ADF) and KPSS tests to assess stationarity.
-
Model Selection:
- Exploration of Auto Regressive, Sarima, and Arima models for forecasting.
-
Forecasting:
- Implementation of selected forecasting models, evaluation using metrics (MAE, RMSE).
- Utilization of Power BI interactive dashboards and reports for effective communication.
- Statistical analysis reveals a significant impact of external events (e.g., earthquake) on sales.
- The earthquake led to a temporary disruption, with sustained higher sales levels for a few months.
- Visual evidence and statistical tests support the conclusion.