behavior_analyzer.py

# behavior_analyzer.py

from dataclasses import dataclass
from typing import List, Dict
import numpy as np


@dataclass
class BehaviorFeatures:
    num_clicks: int
    avg_interval: float
    std_interval: float


class BehaviorAnalyzer:
    """
    Simple heuristic-based analyzer for mouse click behavior.
    It uses:
    - number of clicks
    - average time between clicks
    - variability (std) between clicks
    """

    @staticmethod
    def extract_features(click_timestamps: List[float]) -> BehaviorFeatures:
        if len(click_timestamps) < 2:
            return BehaviorFeatures(
                num_clicks=len(click_timestamps),
                avg_interval=0.0,
                std_interval=0.0,
            )

        diffs = np.diff(sorted(click_timestamps))
        avg_interval = float(np.mean(diffs))
        std_interval = float(np.std(diffs))

        return BehaviorFeatures(
            num_clicks=len(click_timestamps),
            avg_interval=avg_interval,
            std_interval=std_interval,
        )

    @staticmethod
    def compute_score(features: BehaviorFeatures) -> float:
        """
        Returns a score between 0 and 1.
        Higher => more likely to be human.
        """

        # Not enough data
        if features.num_clicks < 5:
            return 0.2

        score = 0.3  # base

        # Reasonable average interval (not too perfect, not too slow)
        if 0.15 <= features.avg_interval <= 1.5:
            score += 0.4

        # Some variability in timing (humans are not perfectly constant)
        if 0.05 <= features.std_interval <= 0.6:
            score += 0.3

        # Clamp between 0 and 1
        score = max(0.0, min(1.0, score))
        return score

    @staticmethod
    def describe(features: BehaviorFeatures, score: float) -> Dict:
        return {
            "num_clicks": features.num_clicks,
            "avg_interval_seconds": round(features.avg_interval, 3),
            "std_interval_seconds": round(features.std_interval, 3),
            "behavior_score": round(score, 3),
        }