Skip to content
Natnael Masresha Zerihun

Pharmaceutical Sales Forecasting

10 Academy · 2022 · Academic project · Individual

End-to-end ML forecasting of store sales six weeks ahead from promotions, competition, holidays, and seasonality, deployed as an interactive Streamlit dashboard.

  • Time series
  • LSTM
  • Random Forest
  • Streamlit
  • DVC

Overview

A time-series forecasting solution on the Rossmann retail dataset: feature engineering across 25+ features, Random Forest and LSTM models, DVC-versioned data, and a Streamlit dashboard where finance teams can get forecasts from manual inputs or CSV uploads.

Media

Bar chart of feature importances
Random-forest feature importance: customer count, store openings, and promotions dominate.
Correlation heatmap of dataset features
Correlation structure of the engineered features.
Histogram of competition distance
Competition-distance distribution motivating the skewness correction during preprocessing.
Training and validation loss curves
LSTM training and validation loss over 200 epochs.

Resources