9/ Explainable AI

Dash integrates Explainable AI (XAI) principles to ensure users understand the reasoning and logic behind every prediction, alert, and decision made by the platform. This transparency enhances trust and empowers users to make better-informed decisions in a complex trading environment.

1. Feature Contribution Analysis

Dash leverages SHAP (SHapley Additive exPlanations) values to quantify the impact of individual metrics on predictions. This allows users to see exactly how each factor contributes to a flagged alert or recommendation.

• SHAP Formula: The contribution of a feature xix_ixi​ to a prediction is calculated as:

  ϕi=∑S⊆N∖{i}∣S∣!(∣N∣−∣S∣−1)!∣N∣!⋅[f(S∪{i})−f(S)]\phi_i = \sum_{S \subseteq N \setminus \{i\}} \frac{|S|! (|N| - |S| - 1)!}{|N|!} \cdot \left[f(S \cup \{i\}) - f(S)\right]ϕi​=S⊆N∖{i}∑​∣N∣!∣S∣!(∣N∣−∣S∣−1)!​⋅[f(S∪{i})−f(S)]

  Where:

  • NNN: Total set of features (e.g., liquidity, trading volume).

  • SSS: Subset of features excluding xix_ixi​.

  • f(S)f(S)f(S): Model output based on subset SSS.

  Example: If the prediction for a token’s high-risk score is driven by 80% liquidity drop and 20% whale sell-off, SHAP values assign these weights explicitly.

2. Traceable Pipelines

Dash ensures full traceability of its decision-making processes, linking every alert to its underlying data sources and metrics.

• Auditability: Each alert provides:

  • The original blockchain data (e.g., Sonic block ID).

  • The exact thresholds that triggered the alert (e.g., Ts>50T_{s} > 50Ts​>50).

• Data Provenance: Alerts are enriched with metadata showing:

  • Data ingestion time (e.g., 10ms ago).

  • Source nodes (e.g., Raydium API for DEX data).

3. Confidence Scoring

Every prediction or alert is accompanied by a confidence score that quantifies the reliability of the AI’s decision.

• Confidence Score Formula:

  Cs=Correct PredictionsTotal Predictions⋅100C_{s} = \frac{\text{Correct Predictions}}{\text{Total Predictions}} \cdot 100Cs​=Total PredictionsCorrect Predictions​⋅100

  Example: If Dash AI accurately predicted 95 out of 100 token trends:

  Cs=95100⋅100=95%C_{s} = \frac{95}{100} \cdot 100 = 95\%Cs​=10095​⋅100=95%

• Multi-Factor Confidence: Scores are influenced by:

  • Data consistency (e.g., high confidence for stable DEX data).

  • Model accuracy (e.g., historical performance of predictions).

  • Temporal factors (e.g., recency of blockchain data).

4. Risk Explanation Breakdown

For every token flagged as high risk, Dash provides a detailed breakdown of the contributing factors.

• Example Risk Report:

  • Token Name: ExampleCoin

  • Risk Score: 85%

  • Contributing Factors:

    • Liquidity Drop: 60% weight

    • Whale Sell-Off: 25% weight

    • Contract Vulnerability: 15% weight

  Users can view this breakdown directly within the platform for enhanced clarity.

5. Visual Interpretability

Dash dashboard includes interactive visualizations to help users interpret predictions intuitively.

• Heatmaps: Show the relative importance of features across multiple tokens.

• Trend Graphs: Overlay predicted trends with actual market movements for validation.

• Anomaly Detection Visuals: Highlight unusual wallet or market behaviors in an easy-to-understand format.

6. Continuous Learning and Feedback

Dash AI models evolve with user feedback to improve prediction accuracy and transparency.

• User Feedback Loop:

  • Users can rate the accuracy of alerts and predictions.

  • Feedback is incorporated into retraining models.

• Real-Time Updates:

  • Models are retrained every 24 hours to adapt to the latest market conditions.