Stage 1

Direction edge experiment ?

Back to Experiments

Experiment Summary

Random Forest ยท 60s direction target
completed
Train rows
15797
Test rows
6771
Default confidence
0.7
Accuracy
0.3843
Progress stage
-
Started
2026-07-04 21:47:18.597086+02:00
Heartbeat
-
Completed
2026-07-04 21:56:18.869420+02:00

Training Dataset Diagnostics

Eligibility funnel captured when this experiment trained.
Quality >= 60
Feature snapshots
111580
Database at training
Selected horizon labels
111580
60s horizon
After quality filter
22568
Eligible label rows
Final dataframe
22568
Rows before 70/30 split
Total snapshots Total labels Labeled labels Labels joined Labeled joined Quality rows Distinct snapshots Duplicate labels
111580 111580 22596 111580 22596 22568 22568 0
Training min captured_at
2026-06-21T16:19:15.790221+02:00
Training max captured_at
2026-06-24T07:11:09.098465+02:00
Latest snapshot at training
2026-07-04T21:47:42.461650+02:00
Latest snapshot now
-
Filters applied: horizon_seconds == 60, is_labeled == true, direction_label is not null, future_return_percent is not null, data_quality_score >= 60. Not applied: no hardcoded row limit, no captured_at date range filter, no training_session_id/session filter.

Class Metrics

ClassPrecisionRecallF1
long 0.3543 0.5138 0.4194
short 0.354 0.4118 0.3807
flat 0.593 0.2262 0.3275

Confusion Matrix

Actual \ PredictedLongShortFlat
Long1157914181
Short1173938167
Flat936798507

Feature Importance

Top 20 features reported by the trained model.
spread_percent
0.024992
cross_exchange_spread_percent
0.024899
cross_exchange_price_deviation_percent
0.024764
mid_price_delta
0.024521
ask_wall_score
0.024016
candle_momentum_5
0.023942
candle_return_3
0.02362
wall_shift_score
0.02335
imbalance_5_delta
0.022615
capture_latency_ms
0.022453
imbalance_10_delta
0.022247
order_book_imbalance_10
0.02222
bid_wall_score
0.021759
order_book_imbalance_5
0.021644
volume_24h
0.021602
order_book_imbalance_20
0.021589
candle_volume_avg_5
0.02071
buy_sell_ratio
0.02064
bid_depth_10_delta
0.020567
ask_depth_10_delta
0.020479

Probability Diagnostics

Maximum class probabilities on the held-out rows used by shadow backtesting.

Probability diagnostics unavailable

No labeled held-out rows remain for diagnostics.

Run Shadow Backtest

Historical simulation on held-out test rows only. No orders or live predictions.

Shadow Backtest Results

Expectancy values are percentages per accepted historical signal.
EdgeConfidenceSignalsLong / ShortWin rate % Avg win %Avg loss %Raw expectancy %Net expectancy % Max drawdown %Profit factorFees + slip %Created
Weak / Insufficient 0.7 1 0 / 1 100 2.0382 - 2.1882 2.0382 0 - 0.15 2026-07-04 21:56