File size: 2,419 Bytes
026c2fb
 
69a9f6c
 
 
 
 
 
 
 
 
026c2fb
69a9f6c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
---
license: mit
task_categories:
  - tabular-classification
  - audio-classification
tags:
  - fraud-detection
  - finance
  - synthetic
  - multi-modal
pretty_name: Fraud Detector Dataset
---

# Fraud Detector Dataset

Synthetic multi-modal dataset created for the **Reply / ReplyMirror Fraud Detection Challenge**. It simulates real-world banking activity across five data types — transactions, users, locations, communications, and audio calls — designed to support fraud detection research combining tabular, textual, geospatial, and audio signals.

The companion system that consumes this data is available at [Honi05/Fraud-Detector](https://github.com/Honi05/Fraud-Detector).

---

## Dataset Structure

`
FraudDetector/
├── transactions.csv
├── users.json
├── locations.json
├── sms.json
├── mails.json
└── audio/
    └── *.mp3  (48 files)
`

---

## File Descriptions

### 	ransactions.csv
The core file. Contains individual financial transactions with fields for transaction ID, amount, timestamp, merchant, and other contextual attributes. This is the primary target file — fraud predictions are made at the transaction level.

### users.json
User profile records. Includes demographic and account information for each customer. Used to build behavioral baselines and detect deviations from a user's normal activity patterns.

### locations.json
Geolocation records linked to transactions or user activity. Used to flag geographical inconsistencies such as impossible travel sequences or transactions in unusual regions.

### sms.json
SMS message logs associated with users. Analyzed for phishing indicators, suspicious links, and fraud-related language patterns using LLM-based text scoring.

### mails.json
Email logs per user. Similar to SMS — processed for social engineering cues, fraud language, and anomalous communication behavior.

### udio/ (48 MP3 files)
Recorded phone call segments named by timestamp and participant (e.g., 20870117_010505-jolanda_orsini.mp3). Intended for voice-based fraud signal extraction.

---

## Usage

This dataset is used as input to a multi-agent fraud detection pipeline:

`ash
python main.py --data ./data --output ./output/predictions.txt
`

The pipeline ingests all five modalities and produces a ranked list of flagged transaction IDs.

---

## Team

**Masala Techii** — Owais Mehboob, Sanya Khan, Honi Arora