Neural Forecast
- Indicators
-
Agus Santoso
DISCLAIMER ON : The author of EA will not be held responsible for your losses or problems of any kind if the EA is directly or indirectly responsible for any losses.
Our software is provided “as is”, without any warranty or condition of any kind, express, implied or statutory. - Version: 1.0
- Activations: 5
MQL5 Blogs : https://www.mql5.com/en/blogs/post/773904
MT4 Version : https://www.mql5.com/en/market/product/189775
MT5 Version : https://www.mql5.com/en/market/product/189776
NEURAL FORECAST [tambangEA] is a multi-symbol neural-network forecasting indicator designed for Gold and major Forex pairs on the Daily (D1) timeframe.
The indicator combines a neural forecasting engine with a structured decision dashboard to estimate the current Daily candle's potential High, Low and Close, while also providing directional bias, model strength, forecast risk/reward, validation statistics and a clear BUY / SELL / WAIT decision layer.
Rather than displaying a simple arrow or oscillator signal, NEURAL FORECAST is designed as a complete market decision-support framework.
Neural Network Forecast Engine
The forecasting engine uses a 6 → 6 → 6 → 6 → 3 Multi-Layer Perceptron architecture.
The model processes six normalized market features:
5-Day Momentum
Range Position
RSI-50
5-Day Range %
5-Day Body %
Acceleration
The network produces three primary forecasts:
Pred HIGH — expected Daily upper price extreme
Pred LOW — expected Daily lower price extreme
Pred CLOSE — expected Daily closing-price reference
The forecast is calculated from completed historical bars and remains frozen for the active Daily setup.
Gold and Forex Model Architecture
NEURAL FORECAST automatically detects the selected market.
Gold Model
For XAUUSD / GOLD, the indicator uses a dedicated fixed neural-network model with:
147 neural-network parameters
and the complete:
6 → 6 → 6 → 6 → 3 architecture
The Gold model operates as a deterministic inference engine without retraining.
Forex Model
For supported Forex symbols, the indicator uses a separate neural network with the same 147-parameter architecture.
The Forex model can be trained from the selected pair's historical Daily data using:
dedicated training data
validation data
reserved out-of-sample data
Huber loss
L2 regularization
gradient clipping
early stopping
best validation epoch restoration
The Forex model therefore adapts its neural weights to the selected currency pair instead of using Gold weights on Forex markets.
Neural Network Map
The right side of the chart can display an interactive visual representation of the neural network:
6 Inputs
→ Hidden Layer 1
→ Hidden Layer 2
→ Hidden Layer 3
→ 3 Outputs
The network map visualizes:
feature nodes
hidden neuron activation
network connections
output nodes
forecast High
forecast Low
forecast Close
The display automatically adjusts to the available chart area.
The neural map can also be disabled when a cleaner chart is preferred.
