mev-beta/pkg/security/anomaly_detector.go

package security

import (
	"fmt"
	"math"
	"sort"
	"sync"
	"sync/atomic"
	"time"

	"github.com/ethereum/go-ethereum/common"
	"github.com/fraktal/mev-beta/internal/logger"
)

// AnomalyDetector detects unusual patterns in security and transaction data
type AnomalyDetector struct {
	logger         *logger.Logger
	config         *AnomalyConfig
	patterns       map[string]*PatternBaseline
	transactionLog []*TransactionRecord
	mu             sync.RWMutex
	alerts         chan *AnomalyAlert
	running        bool
	stopCh         chan struct{}
	alertCounter   uint64
}

// AnomalyConfig configures the anomaly detection system
type AnomalyConfig struct {
	// Detection thresholds
	ZScoreThreshold    float64 `json:"z_score_threshold"`   // Standard deviations for anomaly
	VolumeThreshold    float64 `json:"volume_threshold"`    // Volume change threshold
	FrequencyThreshold float64 `json:"frequency_threshold"` // Frequency change threshold
	PatternSimilarity  float64 `json:"pattern_similarity"`  // Pattern similarity threshold

	// Time windows
	BaselineWindow  time.Duration `json:"baseline_window"`  // Period for establishing baseline
	DetectionWindow time.Duration `json:"detection_window"` // Period for detecting anomalies
	AlertCooldown   time.Duration `json:"alert_cooldown"`   // Cooldown between similar alerts

	// Data retention
	MaxTransactionHistory int           `json:"max_transaction_history"` // Max transactions to keep
	MaxPatternHistory     int           `json:"max_pattern_history"`     // Max patterns to keep
	CleanupInterval       time.Duration `json:"cleanup_interval"`        // How often to clean old data

	// Feature flags
	EnableVolumeDetection  bool `json:"enable_volume_detection"`  // Enable volume anomaly detection
	EnablePatternDetection bool `json:"enable_pattern_detection"` // Enable pattern anomaly detection
	EnableTimeSeriesAD     bool `json:"enable_time_series_ad"`    // Enable time series anomaly detection
	EnableBehavioralAD     bool `json:"enable_behavioral_ad"`     // Enable behavioral anomaly detection
}

// PatternBaseline represents the baseline pattern for a specific metric
type PatternBaseline struct {
	MetricName       string             `json:"metric_name"`
	Observations     []float64          `json:"observations"`
	Mean             float64            `json:"mean"`
	StandardDev      float64            `json:"standard_dev"`
	Variance         float64            `json:"variance"`
	Min              float64            `json:"min"`
	Max              float64            `json:"max"`
	Percentiles      map[int]float64    `json:"percentiles"` // 50th, 75th, 90th, 95th, 99th
	LastUpdated      time.Time          `json:"last_updated"`
	SampleCount      int64              `json:"sample_count"`
	SeasonalPatterns map[string]float64 `json:"seasonal_patterns"` // hour, day, week patterns
	Trend            float64            `json:"trend"`             // Linear trend coefficient
}

// TransactionRecord represents a transaction for anomaly analysis
type TransactionRecord struct {
	Hash         common.Hash            `json:"hash"`
	From         common.Address         `json:"from"`
	To           *common.Address        `json:"to"`
	Value        float64                `json:"value"`     // In ETH
	GasPrice     float64                `json:"gas_price"` // In Gwei
	GasUsed      uint64                 `json:"gas_used"`
	Timestamp    time.Time              `json:"timestamp"`
	BlockNumber  uint64                 `json:"block_number"`
	Success      bool                   `json:"success"`
	Metadata     map[string]interface{} `json:"metadata"`
	AnomalyScore float64                `json:"anomaly_score"`
	Flags        []string               `json:"flags"`
}

// AnomalyAlert represents an detected anomaly
type AnomalyAlert struct {
	ID              string                 `json:"id"`
	Type            AnomalyType            `json:"type"`
	Severity        AnomalySeverity        `json:"severity"`
	Confidence      float64                `json:"confidence"` // 0-1
	Score           float64                `json:"score"`      // Anomaly score
	Description     string                 `json:"description"`
	MetricName      string                 `json:"metric_name"`
	ObservedValue   float64                `json:"observed_value"`
	ExpectedValue   float64                `json:"expected_value"`
	Deviation       float64                `json:"deviation"` // Z-score or similar
	Timestamp       time.Time              `json:"timestamp"`
	Source          string                 `json:"source"` // IP, address, etc.
	Context         map[string]interface{} `json:"context"`
	Recommendations []string               `json:"recommendations"`
	RelatedAlerts   []string               `json:"related_alerts"`
}

// AnomalyType represents the type of anomaly detected
type AnomalyType string

const (
	AnomalyTypeVolume      AnomalyType = "VOLUME"
	AnomalyTypeFrequency   AnomalyType = "FREQUENCY"
	AnomalyTypePattern     AnomalyType = "PATTERN"
	AnomalyTypeBehavioral  AnomalyType = "BEHAVIORAL"
	AnomalyTypeTemporal    AnomalyType = "TEMPORAL"
	AnomalyTypeStatistical AnomalyType = "STATISTICAL"
)

// AnomalySeverity represents the severity of an anomaly
type AnomalySeverity string

const (
	AnomalySeverityLow      AnomalySeverity = "LOW"
	AnomalySeverityMedium   AnomalySeverity = "MEDIUM"
	AnomalySeverityHigh     AnomalySeverity = "HIGH"
	AnomalySeverityCritical AnomalySeverity = "CRITICAL"
)

// NewAnomalyDetector creates a new anomaly detector
func NewAnomalyDetector(logger *logger.Logger, config *AnomalyConfig) *AnomalyDetector {
	cfg := defaultAnomalyConfig()

	if config != nil {
		if config.ZScoreThreshold > 0 {
			cfg.ZScoreThreshold = config.ZScoreThreshold
		}
		if config.VolumeThreshold > 0 {
			cfg.VolumeThreshold = config.VolumeThreshold
		}
		if config.FrequencyThreshold > 0 {
			cfg.FrequencyThreshold = config.FrequencyThreshold
		}
		if config.PatternSimilarity > 0 {
			cfg.PatternSimilarity = config.PatternSimilarity
		}
		if config.BaselineWindow > 0 {
			cfg.BaselineWindow = config.BaselineWindow
		}
		if config.DetectionWindow > 0 {
			cfg.DetectionWindow = config.DetectionWindow
		}
		if config.AlertCooldown > 0 {
			cfg.AlertCooldown = config.AlertCooldown
		}
		if config.MaxTransactionHistory > 0 {
			cfg.MaxTransactionHistory = config.MaxTransactionHistory
		}
		if config.MaxPatternHistory > 0 {
			cfg.MaxPatternHistory = config.MaxPatternHistory
		}
		if config.CleanupInterval > 0 {
			cfg.CleanupInterval = config.CleanupInterval
		}

		cfg.EnableVolumeDetection = config.EnableVolumeDetection
		cfg.EnablePatternDetection = config.EnablePatternDetection
		cfg.EnableTimeSeriesAD = config.EnableTimeSeriesAD
		cfg.EnableBehavioralAD = config.EnableBehavioralAD
	}

	ad := &AnomalyDetector{
		logger:         logger,
		config:         cfg,
		patterns:       make(map[string]*PatternBaseline),
		transactionLog: make([]*TransactionRecord, 0),
		alerts:         make(chan *AnomalyAlert, 1000),
		stopCh:         make(chan struct{}),
	}

	return ad
}

func defaultAnomalyConfig() *AnomalyConfig {
	return &AnomalyConfig{
		ZScoreThreshold:        2.5,
		VolumeThreshold:        3.0,
		FrequencyThreshold:     2.0,
		PatternSimilarity:      0.8,
		BaselineWindow:         24 * time.Hour,
		DetectionWindow:        time.Hour,
		AlertCooldown:          5 * time.Minute,
		MaxTransactionHistory:  10000,
		MaxPatternHistory:      1000,
		CleanupInterval:        time.Hour,
		EnableVolumeDetection:  true,
		EnablePatternDetection: true,
		EnableTimeSeriesAD:     true,
		EnableBehavioralAD:     true,
	}
}

// Start begins the anomaly detection process
func (ad *AnomalyDetector) Start() error {
	ad.mu.Lock()
	defer ad.mu.Unlock()

	if ad.running {
		return nil
	}

	ad.running = true
	go ad.detectionLoop()
	go ad.cleanupLoop()

	ad.logger.Info("Anomaly detector started")
	return nil
}

// Stop stops the anomaly detection process
func (ad *AnomalyDetector) Stop() error {
	ad.mu.Lock()
	defer ad.mu.Unlock()

	if !ad.running {
		return nil
	}

	ad.running = false
	close(ad.stopCh)

	ad.logger.Info("Anomaly detector stopped")
	return nil
}

// RecordTransaction records a transaction for anomaly analysis
func (ad *AnomalyDetector) RecordTransaction(record *TransactionRecord) {
	ad.mu.Lock()
	defer ad.mu.Unlock()

	// Add timestamp if not set
	if record.Timestamp.IsZero() {
		record.Timestamp = time.Now()
	}

	// Calculate initial anomaly score
	record.AnomalyScore = ad.calculateTransactionAnomalyScore(record)

	ad.transactionLog = append(ad.transactionLog, record)

	// Limit history size
	if len(ad.transactionLog) > ad.config.MaxTransactionHistory {
		ad.transactionLog = ad.transactionLog[len(ad.transactionLog)-ad.config.MaxTransactionHistory:]
	}

	// Update patterns
	ad.updatePatternsForTransaction(record)

	// Check for immediate anomalies
	if anomalies := ad.detectTransactionAnomalies(record); len(anomalies) > 0 {
		for _, anomaly := range anomalies {
			select {
			case ad.alerts <- anomaly:
			default:
				ad.logger.Warn("Anomaly alert channel full, dropping alert")
			}
		}
	}
}

// RecordMetric records a metric value for baseline establishment
func (ad *AnomalyDetector) RecordMetric(metricName string, value float64) {
	ad.mu.Lock()
	defer ad.mu.Unlock()

	pattern, exists := ad.patterns[metricName]
	if !exists {
		pattern = &PatternBaseline{
			MetricName:       metricName,
			Observations:     make([]float64, 0),
			Percentiles:      make(map[int]float64),
			SeasonalPatterns: make(map[string]float64),
			LastUpdated:      time.Now(),
		}
		ad.patterns[metricName] = pattern
	}

	// Add observation
	pattern.Observations = append(pattern.Observations, value)
	pattern.SampleCount++

	// Limit observation history
	maxObservations := ad.config.MaxPatternHistory
	if len(pattern.Observations) > maxObservations {
		pattern.Observations = pattern.Observations[len(pattern.Observations)-maxObservations:]
	}

	// Update statistics
	ad.updatePatternStatistics(pattern)

	// Check for anomalies
	if anomaly := ad.detectMetricAnomaly(metricName, value, pattern); anomaly != nil {
		select {
		case ad.alerts <- anomaly:
		default:
			ad.logger.Warn("Anomaly alert channel full, dropping alert")
		}
	}
}

// GetAlerts returns the alert channel for consuming anomaly alerts
func (ad *AnomalyDetector) GetAlerts() <-chan *AnomalyAlert {
	return ad.alerts
}

// GetAnomalyReport generates a comprehensive anomaly report
func (ad *AnomalyDetector) GetAnomalyReport() *AnomalyReport {
	ad.mu.RLock()
	defer ad.mu.RUnlock()

	report := &AnomalyReport{
		Timestamp:            time.Now(),
		PatternsTracked:      len(ad.patterns),
		TransactionsAnalyzed: len(ad.transactionLog),
		AnomaliesDetected:    ad.countRecentAnomalies(),
		TopAnomalies:         ad.getTopAnomalies(10),
		PatternSummaries:     ad.getPatternSummaries(),
		SystemHealth:         ad.calculateSystemHealth(),
	}

	return report
}

// AnomalyReport provides a comprehensive view of anomaly detection status
type AnomalyReport struct {
	Timestamp            time.Time                  `json:"timestamp"`
	PatternsTracked      int                        `json:"patterns_tracked"`
	TransactionsAnalyzed int                        `json:"transactions_analyzed"`
	AnomaliesDetected    int                        `json:"anomalies_detected"`
	TopAnomalies         []*AnomalyAlert            `json:"top_anomalies"`
	PatternSummaries     map[string]*PatternSummary `json:"pattern_summaries"`
	SystemHealth         *AnomalyDetectorHealth     `json:"system_health"`
}

// PatternSummary provides a summary of a pattern baseline
type PatternSummary struct {
	MetricName      string    `json:"metric_name"`
	Mean            float64   `json:"mean"`
	StandardDev     float64   `json:"standard_dev"`
	SampleCount     int64     `json:"sample_count"`
	LastUpdated     time.Time `json:"last_updated"`
	RecentAnomalies int       `json:"recent_anomalies"`
	Trend           string    `json:"trend"` // INCREASING, DECREASING, STABLE
}

// AnomalyDetectorHealth represents the health of the anomaly detection system
type AnomalyDetectorHealth struct {
	IsRunning         bool      `json:"is_running"`
	AlertChannelSize  int       `json:"alert_channel_size"`
	ProcessingLatency float64   `json:"processing_latency_ms"`
	MemoryUsage       int64     `json:"memory_usage_bytes"`
	LastProcessedTime time.Time `json:"last_processed_time"`
	ErrorRate         float64   `json:"error_rate"`
	OverallHealth     string    `json:"overall_health"`
}

// detectionLoop runs the main anomaly detection loop
func (ad *AnomalyDetector) detectionLoop() {
	ticker := time.NewTicker(ad.config.DetectionWindow)
	defer ticker.Stop()

	for {
		select {
		case <-ticker.C:
			ad.performPeriodicDetection()
		case <-ad.stopCh:
			return
		}
	}
}

// cleanupLoop periodically cleans up old data
func (ad *AnomalyDetector) cleanupLoop() {
	ticker := time.NewTicker(ad.config.CleanupInterval)
	defer ticker.Stop()

	for {
		select {
		case <-ticker.C:
			ad.cleanup()
		case <-ad.stopCh:
			return
		}
	}
}

// calculateTransactionAnomalyScore calculates an anomaly score for a transaction
func (ad *AnomalyDetector) calculateTransactionAnomalyScore(record *TransactionRecord) float64 {
	score := 0.0

	// Check value anomaly
	if pattern, exists := ad.patterns["transaction_value"]; exists {
		zScore := ad.calculateZScore(record.Value, pattern)
		score += math.Abs(zScore) * 0.3
	}

	// Check gas price anomaly
	if pattern, exists := ad.patterns["gas_price"]; exists {
		zScore := ad.calculateZScore(record.GasPrice, pattern)
		score += math.Abs(zScore) * 0.2
	}

	// Check frequency anomaly for the sender
	senderFreq := ad.calculateSenderFrequency(record.From)
	if pattern, exists := ad.patterns["sender_frequency"]; exists {
		zScore := ad.calculateZScore(senderFreq, pattern)
		score += math.Abs(zScore) * 0.3
	}

	// Check time-based anomaly
	timeScore := ad.calculateTimeAnomalyScore(record.Timestamp)
	score += timeScore * 0.2

	if score == 0 {
		// Provide a small baseline score for initial transactions without history
		score = 0.1
	}

	return score
}

// detectTransactionAnomalies detects anomalies in a transaction
func (ad *AnomalyDetector) detectTransactionAnomalies(record *TransactionRecord) []*AnomalyAlert {
	var anomalies []*AnomalyAlert

	// Volume anomaly
	if ad.config.EnableVolumeDetection {
		if alert := ad.detectVolumeAnomaly(record); alert != nil {
			anomalies = append(anomalies, alert)
		}
	}

	// Behavioral anomaly
	if ad.config.EnableBehavioralAD {
		if alert := ad.detectBehavioralAnomaly(record); alert != nil {
			anomalies = append(anomalies, alert)
		}
	}

	// Pattern anomaly
	if ad.config.EnablePatternDetection {
		if alert := ad.detectPatternAnomaly(record); alert != nil {
			anomalies = append(anomalies, alert)
		}
	}

	return anomalies
}

// updatePatternsForTransaction updates pattern baselines based on transaction
func (ad *AnomalyDetector) updatePatternsForTransaction(record *TransactionRecord) {
	// Update transaction value pattern
	ad.updatePattern("transaction_value", record.Value)

	// Update gas price pattern
	ad.updatePattern("gas_price", record.GasPrice)

	// Update gas used pattern
	ad.updatePattern("gas_used", float64(record.GasUsed))

	// Update sender frequency
	senderFreq := ad.calculateSenderFrequency(record.From)
	ad.updatePattern("sender_frequency", senderFreq)

	// Update hourly transaction count
	hour := record.Timestamp.Hour()
	hourlyCount := ad.calculateHourlyTransactionCount(hour)
	ad.updatePattern("hourly_transactions", hourlyCount)
}

// updatePattern updates a pattern baseline with a new observation
func (ad *AnomalyDetector) updatePattern(metricName string, value float64) {
	pattern, exists := ad.patterns[metricName]
	if !exists {
		pattern = &PatternBaseline{
			MetricName:       metricName,
			Observations:     make([]float64, 0),
			Percentiles:      make(map[int]float64),
			SeasonalPatterns: make(map[string]float64),
			LastUpdated:      time.Now(),
		}
		ad.patterns[metricName] = pattern
	}

	pattern.Observations = append(pattern.Observations, value)
	pattern.SampleCount++
	pattern.LastUpdated = time.Now()

	// Limit observations
	maxObs := ad.config.MaxPatternHistory
	if len(pattern.Observations) > maxObs {
		pattern.Observations = pattern.Observations[len(pattern.Observations)-maxObs:]
	}

	ad.updatePatternStatistics(pattern)
}

// updatePatternStatistics updates statistical measures for a pattern
func (ad *AnomalyDetector) updatePatternStatistics(pattern *PatternBaseline) {
	if len(pattern.Observations) == 0 {
		return
	}

	// Calculate mean
	sum := 0.0
	for _, obs := range pattern.Observations {
		sum += obs
	}
	pattern.Mean = sum / float64(len(pattern.Observations))

	// Calculate variance and standard deviation
	variance := 0.0
	for _, obs := range pattern.Observations {
		variance += math.Pow(obs-pattern.Mean, 2)
	}
	pattern.Variance = variance / float64(len(pattern.Observations))
	pattern.StandardDev = math.Sqrt(pattern.Variance)

	// Calculate min and max
	pattern.Min = pattern.Observations[0]
	pattern.Max = pattern.Observations[0]
	for _, obs := range pattern.Observations {
		if obs < pattern.Min {
			pattern.Min = obs
		}
		if obs > pattern.Max {
			pattern.Max = obs
		}
	}

	// Calculate percentiles
	sortedObs := make([]float64, len(pattern.Observations))
	copy(sortedObs, pattern.Observations)
	sort.Float64s(sortedObs)

	percentiles := []int{50, 75, 90, 95, 99}
	for _, p := range percentiles {
		index := int(float64(p)/100.0*float64(len(sortedObs)-1) + 0.5)
		if index >= len(sortedObs) {
			index = len(sortedObs) - 1
		}
		pattern.Percentiles[p] = sortedObs[index]
	}

	// Calculate trend (simple linear regression)
	pattern.Trend = ad.calculateTrend(pattern.Observations)
}

// calculateZScore calculates the Z-score for a value against a pattern
func (ad *AnomalyDetector) calculateZScore(value float64, pattern *PatternBaseline) float64 {
	if pattern.StandardDev == 0 {
		return 0
	}
	return (value - pattern.Mean) / pattern.StandardDev
}

// detectMetricAnomaly detects anomalies in a metric value
func (ad *AnomalyDetector) detectMetricAnomaly(metricName string, value float64, pattern *PatternBaseline) *AnomalyAlert {
	if len(pattern.Observations) < 10 { // Need sufficient baseline
		return nil
	}

	zScore := ad.calculateZScore(value, pattern)
	if math.Abs(zScore) < ad.config.ZScoreThreshold {
		return nil
	}

	severity := ad.calculateSeverity(math.Abs(zScore))
	confidence := ad.calculateConfidence(math.Abs(zScore), len(pattern.Observations))

	return &AnomalyAlert{
		ID:            ad.generateAlertID(),
		Type:          AnomalyTypeStatistical,
		Severity:      severity,
		Confidence:    confidence,
		Score:         math.Abs(zScore),
		Description:   ad.generateAnomalyDescription(metricName, value, pattern, zScore),
		MetricName:    metricName,
		ObservedValue: value,
		ExpectedValue: pattern.Mean,
		Deviation:     zScore,
		Timestamp:     time.Now(),
		Context: map[string]interface{}{
			"standard_dev":  pattern.StandardDev,
			"sample_count":  pattern.SampleCount,
			"percentile_95": pattern.Percentiles[95],
		},
		Recommendations: ad.generateRecommendations(metricName, zScore),
	}
}

// Helper methods for specific anomaly types

func (ad *AnomalyDetector) detectVolumeAnomaly(record *TransactionRecord) *AnomalyAlert {
	if record.Value < 0.1 { // Skip small transactions
		return nil
	}

	pattern, exists := ad.patterns["transaction_value"]
	if !exists || len(pattern.Observations) < 10 {
		return nil
	}

	zScore := ad.calculateZScore(record.Value, pattern)
	if math.Abs(zScore) < ad.config.VolumeThreshold {
		return nil
	}

	return &AnomalyAlert{
		ID:            ad.generateAlertID(),
		Type:          AnomalyTypeVolume,
		Severity:      ad.calculateSeverity(math.Abs(zScore)),
		Confidence:    ad.calculateConfidence(math.Abs(zScore), len(pattern.Observations)),
		Score:         math.Abs(zScore),
		Description:   "Unusual transaction volume detected",
		MetricName:    "transaction_value",
		ObservedValue: record.Value,
		ExpectedValue: pattern.Mean,
		Deviation:     zScore,
		Timestamp:     time.Now(),
		Source:        record.From.Hex(),
		Context: map[string]interface{}{
			"transaction_hash": record.Hash.Hex(),
			"gas_price":        record.GasPrice,
			"gas_used":         record.GasUsed,
		},
	}
}

func (ad *AnomalyDetector) detectBehavioralAnomaly(record *TransactionRecord) *AnomalyAlert {
	// Check if sender has unusual behavior
	recentTxs := ad.getRecentTransactionsFrom(record.From, time.Hour)
	if len(recentTxs) < 5 { // Need sufficient history
		return nil
	}

	// Calculate average gas price for this sender
	avgGasPrice := ad.calculateAverageGasPrice(recentTxs)
	gasDeviation := math.Abs(record.GasPrice-avgGasPrice) / avgGasPrice

	if gasDeviation > 2.0 { // 200% deviation
		return &AnomalyAlert{
			ID:            ad.generateAlertID(),
			Type:          AnomalyTypeBehavioral,
			Severity:      AnomalySeverityMedium,
			Confidence:    0.8,
			Score:         gasDeviation,
			Description:   "Unusual gas price behavior for sender",
			MetricName:    "sender_gas_behavior",
			ObservedValue: record.GasPrice,
			ExpectedValue: avgGasPrice,
			Deviation:     gasDeviation,
			Timestamp:     time.Now(),
			Source:        record.From.Hex(),
		}
	}

	return nil
}

func (ad *AnomalyDetector) detectPatternAnomaly(record *TransactionRecord) *AnomalyAlert {
	// Time-based pattern analysis
	hour := record.Timestamp.Hour()
	hourlyPattern, exists := ad.patterns["hourly_transactions"]
	if !exists {
		return nil
	}

	// Check if this hour is unusual for transaction activity
	hourlyCount := ad.calculateHourlyTransactionCount(hour)
	zScore := ad.calculateZScore(hourlyCount, hourlyPattern)

	if math.Abs(zScore) > ad.config.PatternSimilarity*2 {
		return &AnomalyAlert{
			ID:            ad.generateAlertID(),
			Type:          AnomalyTypePattern,
			Severity:      ad.calculateSeverity(math.Abs(zScore)),
			Confidence:    0.7,
			Score:         math.Abs(zScore),
			Description:   "Unusual transaction timing pattern",
			MetricName:    "hourly_pattern",
			ObservedValue: hourlyCount,
			ExpectedValue: hourlyPattern.Mean,
			Deviation:     zScore,
			Timestamp:     time.Now(),
			Context: map[string]interface{}{
				"hour": hour,
			},
		}
	}

	return nil
}

// Helper calculation methods

func (ad *AnomalyDetector) calculateSenderFrequency(sender common.Address) float64 {
	count := 0
	cutoff := time.Now().Add(-time.Hour)

	for _, tx := range ad.transactionLog {
		if tx.From == sender && tx.Timestamp.After(cutoff) {
			count++
		}
	}

	return float64(count)
}

func (ad *AnomalyDetector) calculateTimeAnomalyScore(timestamp time.Time) float64 {
	hour := timestamp.Hour()

	// Business hours (9 AM - 5 PM) are normal, night hours are more suspicious
	if hour >= 9 && hour <= 17 {
		return 0.0
	} else if hour >= 22 || hour <= 6 {
		return 0.8 // High suspicion for very late/early hours
	}

	return 0.3 // Medium suspicion for evening hours
}

func (ad *AnomalyDetector) calculateHourlyTransactionCount(hour int) float64 {
	count := 0
	now := time.Now()
	startOfHour := time.Date(now.Year(), now.Month(), now.Day(), hour, 0, 0, 0, now.Location())
	endOfHour := startOfHour.Add(time.Hour)

	for _, tx := range ad.transactionLog {
		if tx.Timestamp.After(startOfHour) && tx.Timestamp.Before(endOfHour) {
			count++
		}
	}

	return float64(count)
}

func (ad *AnomalyDetector) getRecentTransactionsFrom(sender common.Address, duration time.Duration) []*TransactionRecord {
	var result []*TransactionRecord
	cutoff := time.Now().Add(-duration)

	for _, tx := range ad.transactionLog {
		if tx.From == sender && tx.Timestamp.After(cutoff) {
			result = append(result, tx)
		}
	}

	return result
}

func (ad *AnomalyDetector) calculateAverageGasPrice(transactions []*TransactionRecord) float64 {
	if len(transactions) == 0 {
		return 0
	}

	sum := 0.0
	for _, tx := range transactions {
		sum += tx.GasPrice
	}

	return sum / float64(len(transactions))
}

func (ad *AnomalyDetector) calculateTrend(observations []float64) float64 {
	if len(observations) < 2 {
		return 0
	}

	// Simple linear regression
	n := float64(len(observations))
	sumX, sumY, sumXY, sumX2 := 0.0, 0.0, 0.0, 0.0

	for i, y := range observations {
		x := float64(i)
		sumX += x
		sumY += y
		sumXY += x * y
		sumX2 += x * x
	}

	// Calculate slope (trend)
	denominator := n*sumX2 - sumX*sumX
	if denominator == 0 {
		return 0
	}

	return (n*sumXY - sumX*sumY) / denominator
}

func (ad *AnomalyDetector) calculateSeverity(zScore float64) AnomalySeverity {
	if zScore < 2.0 {
		return AnomalySeverityLow
	} else if zScore < 3.0 {
		return AnomalySeverityMedium
	} else if zScore < 4.0 {
		return AnomalySeverityHigh
	}
	return AnomalySeverityCritical
}

func (ad *AnomalyDetector) calculateConfidence(zScore float64, sampleSize int) float64 {
	// Higher Z-score and larger sample size increase confidence
	zConfidence := math.Min(zScore/5.0, 1.0)
	sampleConfidence := math.Min(float64(sampleSize)/100.0, 1.0)
	return (zConfidence + sampleConfidence) / 2.0
}

func (ad *AnomalyDetector) generateAlertID() string {
	counter := atomic.AddUint64(&ad.alertCounter, 1)
	return fmt.Sprintf("anomaly_%d_%d", time.Now().UnixNano(), counter)
}

func (ad *AnomalyDetector) generateAnomalyDescription(metricName string, value float64, pattern *PatternBaseline, zScore float64) string {
	direction := "above"
	if zScore < 0 {
		direction = "below"
	}

	return fmt.Sprintf("Metric '%s' value %.2f is %.1f standard deviations %s the expected mean of %.2f",
		metricName, value, math.Abs(zScore), direction, pattern.Mean)
}

func (ad *AnomalyDetector) generateRecommendations(metricName string, zScore float64) []string {
	recommendations := []string{}

	if math.Abs(zScore) > 4.0 {
		recommendations = append(recommendations, "Immediate investigation required")
		recommendations = append(recommendations, "Consider blocking source if malicious")
	} else if math.Abs(zScore) > 3.0 {
		recommendations = append(recommendations, "Initiate investigation and monitor closely")
		recommendations = append(recommendations, "Review transaction history")
	} else {
		recommendations = append(recommendations, "Continue monitoring")
	}

	switch metricName {
	case "transaction_value":
		recommendations = append(recommendations, "Verify large transaction legitimacy")
	case "gas_price":
		recommendations = append(recommendations, "Check for gas price manipulation")
	case "sender_frequency":
		recommendations = append(recommendations, "Investigate potential automated behavior")
	}

	return recommendations
}

// Status and reporting methods

func (ad *AnomalyDetector) performPeriodicDetection() {
	ad.mu.RLock()
	defer ad.mu.RUnlock()

	// Perform time-series analysis on patterns
	for metricName, pattern := range ad.patterns {
		if ad.config.EnableTimeSeriesAD {
			ad.performTimeSeriesAnalysis(metricName, pattern)
		}
	}

	// Analyze transaction patterns
	ad.analyzeTransactionPatterns()
}

func (ad *AnomalyDetector) performTimeSeriesAnalysis(metricName string, pattern *PatternBaseline) {
	if len(pattern.Observations) < 20 {
		return
	}

	// Look for sudden changes in recent observations
	recentWindow := 5
	if len(pattern.Observations) < recentWindow {
		return
	}

	recent := pattern.Observations[len(pattern.Observations)-recentWindow:]
	historical := pattern.Observations[:len(pattern.Observations)-recentWindow]

	recentMean := ad.calculateMean(recent)
	historicalMean := ad.calculateMean(historical)
	historicalStdDev := ad.calculateStdDev(historical, historicalMean)

	if historicalStdDev > 0 {
		changeScore := math.Abs(recentMean-historicalMean) / historicalStdDev
		if changeScore > ad.config.ZScoreThreshold {
			alert := &AnomalyAlert{
				ID:            ad.generateAlertID(),
				Type:          AnomalyTypeTemporal,
				Severity:      ad.calculateSeverity(changeScore),
				Confidence:    0.9,
				Score:         changeScore,
				Description:   "Significant change in time series pattern detected",
				MetricName:    metricName,
				ObservedValue: recentMean,
				ExpectedValue: historicalMean,
				Deviation:     changeScore,
				Timestamp:     time.Now(),
			}

			select {
			case ad.alerts <- alert:
			default:
			}
		}
	}
}

func (ad *AnomalyDetector) analyzeTransactionPatterns() {
	// Look for unusual transaction patterns
	recentTxs := ad.getRecentTransactions(time.Hour)

	// Analyze sender patterns
	senderCounts := make(map[common.Address]int)
	for _, tx := range recentTxs {
		senderCounts[tx.From]++
	}

	// Check for addresses with unusual frequency
	for sender, count := range senderCounts {
		if count > 50 { // Threshold for suspicious activity
			alert := &AnomalyAlert{
				ID:            ad.generateAlertID(),
				Type:          AnomalyTypeFrequency,
				Severity:      AnomalySeverityHigh,
				Confidence:    0.95,
				Score:         float64(count),
				Description:   "High frequency transaction pattern detected",
				MetricName:    "transaction_frequency",
				ObservedValue: float64(count),
				ExpectedValue: 10.0, // Expected average
				Timestamp:     time.Now(),
				Source:        sender.Hex(),
			}

			select {
			case ad.alerts <- alert:
			default:
			}
		}
	}
}

func (ad *AnomalyDetector) getRecentTransactions(duration time.Duration) []*TransactionRecord {
	var result []*TransactionRecord
	cutoff := time.Now().Add(-duration)

	for _, tx := range ad.transactionLog {
		if tx.Timestamp.After(cutoff) {
			result = append(result, tx)
		}
	}

	return result
}

func (ad *AnomalyDetector) cleanup() {
	ad.mu.Lock()
	defer ad.mu.Unlock()

	// Clean old transaction records
	cutoff := time.Now().Add(-ad.config.BaselineWindow)
	newLog := make([]*TransactionRecord, 0)

	for _, tx := range ad.transactionLog {
		if tx.Timestamp.After(cutoff) {
			newLog = append(newLog, tx)
		}
	}

	ad.transactionLog = newLog

	// Clean old pattern observations
	for _, pattern := range ad.patterns {
		if len(pattern.Observations) > ad.config.MaxPatternHistory {
			pattern.Observations = pattern.Observations[len(pattern.Observations)-ad.config.MaxPatternHistory:]
			ad.updatePatternStatistics(pattern)
		}
	}
}

func (ad *AnomalyDetector) countRecentAnomalies() int {
	// This would count anomalies in the last hour
	// For now, return a placeholder
	return 0
}

func (ad *AnomalyDetector) getTopAnomalies(limit int) []*AnomalyAlert {
	// This would return the top anomalies by score
	// For now, return empty slice
	return []*AnomalyAlert{}
}

func (ad *AnomalyDetector) getPatternSummaries() map[string]*PatternSummary {
	ad.mu.RLock()
	defer ad.mu.RUnlock()

	summaries := make(map[string]*PatternSummary)

	for name, pattern := range ad.patterns {
		trend := "STABLE"
		if pattern.Trend > 0.1 {
			trend = "INCREASING"
		} else if pattern.Trend < -0.1 {
			trend = "DECREASING"
		}

		summaries[name] = &PatternSummary{
			MetricName:      name,
			Mean:            pattern.Mean,
			StandardDev:     pattern.StandardDev,
			SampleCount:     pattern.SampleCount,
			LastUpdated:     pattern.LastUpdated,
			RecentAnomalies: 0, // Would count recent anomalies for this pattern
			Trend:           trend,
		}
	}

	return summaries
}

func (ad *AnomalyDetector) calculateSystemHealth() *AnomalyDetectorHealth {
	ad.mu.RLock()
	defer ad.mu.RUnlock()

	return &AnomalyDetectorHealth{
		IsRunning:         ad.running,
		AlertChannelSize:  len(ad.alerts),
		ProcessingLatency: 5.2,              // Would measure actual latency
		MemoryUsage:       1024 * 1024 * 10, // Would measure actual memory
		LastProcessedTime: time.Now(),
		ErrorRate:         0.0, // Would track actual error rate
		OverallHealth:     "HEALTHY",
	}
}

// Helper calculation functions

func (ad *AnomalyDetector) calculateMean(values []float64) float64 {
	if len(values) == 0 {
		return 0
	}

	sum := 0.0
	for _, v := range values {
		sum += v
	}

	return sum / float64(len(values))
}

func (ad *AnomalyDetector) calculateStdDev(values []float64, mean float64) float64 {
	if len(values) == 0 {
		return 0
	}

	variance := 0.0
	for _, v := range values {
		variance += math.Pow(v-mean, 2)
	}
	variance /= float64(len(values))

	return math.Sqrt(variance)
}