RaSPDAM Dataset Overview

Dataset Introduction

RaSPDAM is a machine‑learning algorithm based on visual morphological features, specifically designed for the detection of Fast Radio Bursts (FRBs). The algorithm addresses the computational and time‑consumption challenges of traditional FRB search methods, offering significant improvements in efficiency and accuracy.

Dataset Background

Fast Radio Bursts (FRBs) are short, intense radio‑energy pulses emitted from distant galaxies. Discovering and studying FRBs is crucial for understanding matter distribution and evolution in the universe. However, the massive data volumes produced by radio telescopes and the computational complexity of existing search methods make FRB detection a demanding task. Conventional techniques often perform poorly on weak signals and are time‑consuming.

Dataset Content

The RaSPDAM test is based on the FAST‑FREX dataset, which is built on observations from FAST. The dataset contains:

• 600 positive samples: observed FRB signals from three sources (FRB20121102, FRB20180301, FRB20201124).
• 1,000 negative samples: noise and radio frequency interference (RFI).

Key Features

Efficiency and Accuracy

• High Precision: RaSPDAM achieves 98.73% precision, far surpassing traditional methods such as PRESTO and Heimdall.
• High Recall: Recall reaches 77.67%, effectively identifying the majority of true FRB signals.
• F1 Score: 0.8694, indicating a good balance between precision and recall.

Versatility

• Currently, RaSPDAM provides Time of Arrival (ToA) as output; future enhancements will include Dispersion Measure (DM) for more comprehensive signal verification.

Performance Benchmark

Comparison with traditional methods:

| Software | TN | TP | FN | FP | Recall | Precision | F1 |
|----------|----|----|----|----|--------|-----------|----|
| PRESTO   | 3  | 472| 0  | 26963700 | 0.7867 | 1.7505E-05 | 3.5009E-05 |
| Heimdall | 218| 489| 36 | 5854    | 0.8150 | 0.0771    | 0.1409 |
| RaSPDAM  | 989| 466| 128| 6       | 0.7767 | 0.9873    | 0.8694 |

Findings

Since deployment, RaSPDAM has played a crucial role in identifying:

• 2 new FRBs: FRB20211103A and FRB20230104.
• 80 pulsars, including 13 previously undiscovered pulsars, highlighting the algorithm's effectiveness in discovering new astronomical objects.