P2P Bolivia ML — Dashboard Profesional

Métricas Clave

RMSE · LSTM

—

Root Mean Squared Error

RMSE · Transformer

—

Root Mean Squared Error

RMSE · Transfer Learning

—

Root Mean Squared Error

R² · Mejor Modelo

—

Coeficiente de determinación

MAPE · LSTM

—

Mean Absolute % Error

Precio P2P

— BOB

Último precio real

Sharpe · LSTM

—

Trading Sharpe Ratio

Pipeline

—

Tiempo total de entrenamiento

Predicciones en Tiempo Real vs Precio Real

Análisis de Modelos

Error de Predicción |Real − Pred|

Comparación de Modelos

LSTM

🏆 Ganador

RMSE—

MAE—

MAPE—

R²—

DA—

Params—

Transformer

2do

RMSE—

MAE—

MAPE—

R²—

DA—

Params—

Transfer Learning

3ro

RMSE—

MAE—

MAPE—

R²—

DA—

Params—

Trading Simulator

Capital inicial: $10,000

Retorno y Sharpe Ratio

Max Drawdown y Trades

Walk-Forward Validation

5 splits · Expanding window

RMSE por Modelo (Walk-Forward)

Directional Accuracy (Walk-Forward)

BTC/USDT Spot + Precio P2P Bolivia (BOB)

Arquitectura del Pipeline

Data Layer

📡

CoinGecko API

📊

Binance Spot

🔄

Binance P2P

📈

70 Features

Preprocessing

🧹

Outlier Removal

📏

StandardScaler

⚙️

Feature Eng.

Model Layer

🧠

Stacked Bi-LSTM
+ Temporal Attention

⚡

Pre-LN Transformer
ProbSparse Attention

🔄

Transfer Learning
BTC → P2P Bolivia

Evaluation

📋

Métricas

📉

Walk-Forward

💰

Trading Sim

📊

Diebold-Mariano

🔬

Lookback

72h

3 días de contexto

🎯

Target

p2p_price_bob

Variable objetivo

📊

Features

70

Variables de entrada

📅

Train/Val/Test

2020–2024

Split temporal

🔢

Secuencias

26,151

Train sequences

⏱️

Diebold-Mariano

p=0.0000

LSTM sig. mejor