process-mining-analyzer

What this skill does

# process-mining-analyzer

You are **process-mining-analyzer** - a specialized skill for process mining including event log analysis, process discovery, and conformance checking.

## Overview

This skill enables AI-powered process mining including:
- Event log preparation and cleaning
- Process discovery algorithms (Alpha, Heuristic Miner)
- Conformance checking
- Performance analysis
- Bottleneck identification
- Variant analysis
- Social network analysis
- Dotted chart visualization

## Capabilities

### 1. Event Log Preparation

```python
import pandas as pd
import numpy as np
from datetime import datetime
from collections import defaultdict

def prepare_event_log(raw_data: pd.DataFrame, mappings: dict):
    """
    Prepare event log for process mining

    raw_data: DataFrame with raw event data
    mappings: {'case_id': col, 'activity': col, 'timestamp': col, 'resource': col}
    """
    # Map columns
    event_log = pd.DataFrame()
    event_log['case_id'] = raw_data[mappings['case_id']]
    event_log['activity'] = raw_data[mappings['activity']]
    event_log['timestamp'] = pd.to_datetime(raw_data[mappings['timestamp']])

    if 'resource' in mappings and mappings['resource'] in raw_data.columns:
        event_log['resource'] = raw_data[mappings['resource']]

    # Sort by case and timestamp
    event_log = event_log.sort_values(['case_id', 'timestamp'])

    # Add derived columns
    event_log['event_id'] = range(len(event_log))

    # Calculate duration to next event
    event_log['next_timestamp'] = event_log.groupby('case_id')['timestamp'].shift(-1)
    event_log['duration'] = (event_log['next_timestamp'] - event_log['timestamp']).dt.total_seconds()

    # Statistics
    stats = {
        'total_events': len(event_log),
        'total_cases': event_log['case_id'].nunique(),
        'unique_activities': event_log['activity'].nunique(),
        'activities': event_log['activity'].unique().tolist(),
        'date_range': {
            'start': str(event_log['timestamp'].min()),
            'end': str(event_log['timestamp'].max())
        }
    }

    return {
        'event_log': event_log,
        'statistics': stats
    }
```

### 2. Process Discovery

```python
def discover_process_model(event_log: pd.DataFrame):
    """
    Discover process model from event log using footprint analysis
    """
    # Build directly-follows graph
    dfg = defaultdict(int)
    start_activities = set()
    end_activities = set()

    for case_id, case_data in event_log.groupby('case_id'):
        activities = case_data['activity'].tolist()

        if activities:
            start_activities.add(activities[0])
            end_activities.add(activities[-1])

        for i in range(len(activities) - 1):
            dfg[(activities[i], activities[i + 1])] += 1

    # Build footprint matrix
    activities = sorted(event_log['activity'].unique())
    n = len(activities)
    act_idx = {a: i for i, a in enumerate(activities)}

    # Relations: > (directly follows), < (preceded by), || (parallel), # (no relation)
    relations = {}

    for a1 in activities:
        for a2 in activities:
            a1_to_a2 = dfg.get((a1, a2), 0)
            a2_to_a1 = dfg.get((a2, a1), 0)

            if a1_to_a2 > 0 and a2_to_a1 > 0:
                relations[(a1, a2)] = '||'  # Parallel
            elif a1_to_a2 > 0:
                relations[(a1, a2)] = '>'   # Follows
            elif a2_to_a1 > 0:
                relations[(a1, a2)] = '<'   # Preceded by
            else:
                relations[(a1, a2)] = '#'   # No relation

    return {
        'directly_follows_graph': dict(dfg),
        'start_activities': list(start_activities),
        'end_activities': list(end_activities),
        'footprint': relations,
        'activities': activities
    }

def heuristic_miner(event_log: pd.DataFrame, dependency_threshold: float = 0.5):
    """
    Heuristic Miner algorithm for process discovery
    """
    # Build frequency tables
    activity_freq = event_log['activity'].value_counts().to_dict()
    dfg = defaultdict(int)

    for case_id, case_data in event_log.groupby('case_id'):
        activities = case_data['activity'].tolist()
        for i in range(len(activities) - 1):
            dfg[(activities[i], activities[i + 1])] += 1

    # Calculate dependency measure
    # D(a,b) = (|a>b| - |b>a|) / (|a>b| + |b>a| + 1)
    dependencies = {}
    activities = list(activity_freq.keys())

    for a in activities:
        for b in activities:
            a_to_b = dfg.get((a, b), 0)
            b_to_a = dfg.get((b, a), 0)

            if a_to_b > 0 or b_to_a > 0:
                dep = (a_to_b - b_to_a) / (a_to_b + b_to_a + 1)
                if abs(dep) >= dependency_threshold:
                    dependencies[(a, b)] = round(dep, 3)

    # Filter to positive dependencies (actual follows relations)
    causal_relations = {k: v for k, v in dependencies.items() if v > 0}

    return {
        'activity_frequencies': activity_freq,
        'directly_follows_frequencies': dict(dfg),
        'dependency_measures': dependencies,
        'causal_relations': causal_relations,
        'threshold': dependency_threshold
    }
```

### 3. Conformance Checking

```python
def check_conformance(event_log: pd.DataFrame, expected_sequence: list,
                     strict: bool = False):
    """
    Check conformance of traces against expected process

    expected_sequence: list of activities in expected order
    strict: if True, exact match required; if False, subsequence match
    """
    results = []

    for case_id, case_data in event_log.groupby('case_id'):
        trace = case_data['activity'].tolist()

        if strict:
            # Exact match
            is_conforming = trace == expected_sequence
            deviations = []

            if not is_conforming:
                # Find deviations
                for i, (actual, expected) in enumerate(zip(trace, expected_sequence)):
                    if actual != expected:
                        deviations.append({
                            'position': i,
                            'expected': expected,
                            'actual': actual
                        })

                # Check for missing or extra activities
                if len(trace) < len(expected_sequence):
                    deviations.append({'type': 'missing', 'count': len(expected_sequence) - len(trace)})
                elif len(trace) > len(expected_sequence):
                    deviations.append({'type': 'extra', 'count': len(trace) - len(expected_sequence)})

        else:
            # Check if expected is subsequence
            exp_idx = 0
            is_conforming = True

            for act in trace:
                if exp_idx < len(expected_sequence) and act == expected_sequence[exp_idx]:
                    exp_idx += 1

            is_conforming = exp_idx == len(expected_sequence)
            deviations = [] if is_conforming else [{'type': 'subsequence_mismatch'}]

        results.append({
            'case_id': case_id,
            'trace': trace,
            'conforming': is_conforming,
            'deviations': deviations
        })

    # Summary statistics
    conforming_count = sum(1 for r in results if r['conforming'])
    total = len(results)

    return {
        'case_results': results,
        'summary': {
            'total_cases': total,
            'conforming_cases': conforming_count,
            'non_conforming_cases': total - conforming_count,
            'conformance_rate': round(conforming_count / total * 100, 1) if total > 0 else 0
        }
    }
```

### 4. Performance Analysis

```python
def analyze_performance(event_log: pd.DataFrame):
    """
    Analyze process performance from event log
    """
    # Case duration
    case_durations = event_log.groupby('case_id').agg({
        'timestamp': ['min', 'max']
    })
    case_durations.columns = ['start', 'end']
    case_durations['duration_hours'] = (case_durations