Demand Forecasting Library
A Python library that forecasts demand from time-series data and, before producing a forecast, evaluates how reliable that data is likely to be.
Duration
Short-term development
Team
1 person
01 // OVERVIEW
Project Overview
A Python library for forecasting demand from time-series data. It forecasts using an ensemble of Prophet and multiple other models, and before issuing a forecast, it estimates likely accuracy from the data's characteristics - a confidence check that shows upfront whether the forecast can be trusted.
02 // KEY FEATURES
Key Features
Prophet + Ensemble Forecasting
In addition to Prophet, provides ensemble forecasting by averaging three models: Theta, ETS, and Seasonal Naive.
Pre-Forecast Confidence Check
Scores expected accuracy from data characteristics and presents it upfront in three levels - reliable, caution, or risky - allowing a warning to be raised before an untrustworthy forecast is issued.
Evaluation Metric Calculation
Evaluates forecast accuracy using metrics such as MAPE, RMSE, and MAE, and performs missing-value imputation, outlier detection, and normalization as preprocessing.
03 // TECHNICAL DEEP-DIVE
Technical Highlights
Solution
Handling Data That Can't Be Reliably Forecast
Forecasting data that is inherently hard to predict accurately, without warning, can lead to poor decisions. A mechanism that judges confidence before forecasting makes clear when the tool should and shouldn't be relied on.
04 // IMPACT & RESULTS
Results
3 Models
Ensemble forecasting
Pre-Forecast Check
Confidence scored upfront