Products & Solutions / By Solution / Forecaster / Feature Set

• Wizard-like interface
• Three interface modes: Basic, Standard and Expert
• Automatic and manual data preprocessing
• Automatic and manual selection of neural network architecture and training parameters
• Detailed reporting
• Online help system
• Free technical support
• Sample financial, marketing, real estate and scientific problems included

• Input dataset size is limited only by the hardware of the computer
• Import popular ASCII file formats (CSV, TXT, PRN) with automatic recognition of delimiter and column headers
• Import Excel files
• Automatic Date/Time values encoding
• Automatic categorical values encoding
• Automatic numeric values scaling
• Missing values handling (removal and 4 substitution options)
• Outliers handling (customizable outlier coefficient)
• Automatic recognition of data entry errors (wrong type values)
• Automatic and manual column type identification (numeric, categorical, date, time, text)
• Automatic random dataset division onto training, validation and test sets
• Detailed Data Analysis Report

For experienced users
• Manual dataset division onto training, validation and test sets (random or sequential)
• Visual representation of data anomalies in Data Analysis Details window
• Ability to accept/ignore rows and columns manually
• Manual min/max values specification to anticipate bigger values in future data for forecasting
• Detailed Preprocessing Report

• Fully automated neural network design - automatically search for the best architecture and train the most suitable neural network to solve your problem
• Three timesaving methods of neural network architecture search
• Exhaustive architecture search with customizable parameters
• Genetic algorithms architecture search with customizable parameters
• Automatic training algorithm selection

For experienced users
• Manual selection of training algorithms: Conjugate Gradient Descent, Levenberg-Marquardt, Quick-Propagation, Incremental and Batch Back-Propagation
• Manual architecture specification (supported network type: multi-layer perceptron)
• Automatic adjustment of learning rate and momentum during training for Back-Propagation algorithm

• Automatic generation of versatile stopping condition to stop network training
• Non-technical presentation of neural network-related options in Standard mode
• Real-time training error graph (absolute error by iteration)
• Early-stopping on generalization loss (10 preset levels)
• Retain and restore best network
• Automatic network retrains and selection of the best network among retrains
• Manual network retrain (4 options)

For experienced users
• Manual stopping conditions (target error level, error improvement, correct classification rate, number of iterations)
• Real-time control on training parameters: errors (MSE, MAE, CCR) on training and validation set, error improvement, training speed, # of iterations

• Estimated forecasting error
• Actual vs Forecasted Table with absolute and relative errors
• Single point forecast and bulk forecasting
• Quick forecast with already trained network

• Detailed explanations on every step
• Customizable reports (with preview and printing capabilities)
• Reports export to HTML and XLS
• Save/Load neural network