🔧 Feature Engineering & Label Design
In modern stock market research, raw price data alone is insufficient for identifying actionable trends. At ApexTrend, we enhance market data using a curated set of technical transformations and labeling logic. These techniques help our models evaluate directional potential over 1 to 3 day windows while filtering short-term noise.
📊 Engineered Input Features
Rather than disclose proprietary formulas, we categorize our feature inputs by theme:
- Technical momentum and mean reversion zones (e.g., RSI ranges, MACD signals)
- Price behavior and structure over rolling time slices (e.g., slopes, multi-candle setups)
- Volume attention and expansion factors (e.g., relative volume, surge scores)
- Trend alignment and strength (e.g., EMA stacking, ADX signals)
- Session-based timing effects (e.g., opening strength, pre-market behaviors)
🎯 Outcome Labeling
To guide our model toward meaningful pattern recognition, we use forward-looking labels to define whether a setup aligned with directional strength:
- Next-day strength (e.g., gain from open to close)
- Intraday follow-through within thresholds (e.g., +2% from prior close)
- Drawdown-limited advances (e.g., price gain without exceeding -1% intraday)
Labels are time-aligned using forward-only indexing to avoid lookahead bias. Our methods respect SEC-compliant research protocols and do not suggest investment advice.
🔐 Designed for Research Integrity
Our system avoids over-disclosure of sensitive formulas or model internals. Instead, we focus on thematic structure and sound research design. This protects the proprietary logic powering ApexTrend while still offering transparency for users seeking a deeper understanding of how our research engine operates.
For a broader overview of how these features power our LightGBM research model, visit our LightGBM Research Engine page.