Product Functionality
This product would be a forecasting sofware designed specifically for business forecasters.The sofware generates accurate forecasts quickly and easily using proven statistical and machine learning algorithms. This sofware builds forecasts by looking at a historical series of data, which is called time series data. Model and algorithms that are used in this sofware are agnostic to industries; meaning it can be used in any industry for any type of time series data. These time series data can be future business outcomes such as product demand, resource needs, or financial performance. The forecasting sofware accommodates seasonal demand, product hierarchies, slow-moving and intermittent demands, causal variables and covariates, outliers, prediction intervals and more. The forecasting sofware automatically analyzes the data, selects the appropriate forecasting technique among all available algorithms, builds models and generates forecasts, all steps are being done programmatically and there is no need to have data scientists to perform those tasks.
AI Platform
It’s a reusable AI platform
Data Lake, Data Processing, Dashboard and Analytics features are general and can be used for different ML like Regression, Classification, Clustering
First AI based set is Time Series.
Support different deployment options(SaaS, PaaS)
Problems to address
The latest AI algorithms and open source AI
Fully automated, designed for business users with a little knowledge of ML and Forecasting
Cloud compatible.
Handle small, medium and large scales
Handle massive data
Solution
We design a platform around well-known algorithms (Microsoft FLAML and H2o Automl)
Add AI new building blocks like Classification, Clustering later
Cloud native solution for private and public clouds
Why now:
AI development has proved that it can provide improvement in time series forecasting over traditional statistical models.
Data availability is no longer an issue and there is enough data to train AI algorithms
Most of the available algorithms still require data scientists to use them, our goal is to make those algorithm user friendly to the average business analysts
Features
Data wrangling and preprocessing
- Data wrangling and preprocessing
- Time series forecasting
- Data post processing and visualization
Time series forecasting
- Machine learning approach by using decision tree family algorithms including lightgbm, xgboost and catboost and
- Machine learning approach by using automated ML algorithms like H2o AutoMl
- Machine learning approach by using neural network algorithms like LSTM
- Statistical approach by using Arima and exponential smoothing algorithms
Data post processing and visualization
- Impose events
- Display accuracy measures per series and grouped series
- Display forecasts graphs per series and grouped series
- Identify and flag potential problematic forecasts and user to review
Market
