When old meets new: evaluating numerical and Machine Learning based eclipse prediction methods

Frincu, M. ORCID: 0000-0003-1034-8409 and Sferdian, M., 2021. When old meets new: evaluating numerical and Machine Learning based eclipse prediction methods. Romanian Astronomical Journal, 31 (2), pp. 133-152. ISSN 1220-5168

[img]
Preview
Text
1465942_Frincu.pdf - Post-print

Download (329kB) | Preview

Abstract

The Babylonians had some of the most advanced arithmetic models for the Lunar and planetary theory in ancient times. This allowed them to discover the Saros eclipse cycle. In this paper we investigate the accuracy of several eclipse prediction methods tested on worldwide occurring eclipses between 2020--2100 CE, ignoring the sophisticated modern models based on accurate ephemerides, in an attempt to understand how simple models would have worked in ancient times. First we propose two numerical methods relying on knowledge available in Babylonian times -- lunar phases, lunar nodes, and the angular separation between the Sun and the Moon. Second, we assess the performance of four Machine Learning (ML) models modeling human inference by relying on the same data. The accuracy of the numerical methods is above 80\% while the ML models achieve up to 98\% accuracy. The algorithms perform better in case of lunar eclipses. While not 100\% perfect, these methods are simplistic in terms of required information and enable us to get an insight into how efficient might have been ancient methods relying on visual observations.

Item Type: Journal article
Publication Title: Romanian Astronomical Journal
Creators: Frincu, M. and Sferdian, M.
Publisher: Romanian Academy Publishing House
Date: 6 September 2021
Volume: 31
Number: 2
ISSN: 1220-5168
Identifiers:
NumberType
1465942Other
Divisions: Schools > School of Science and Technology
Record created by: Linda Sullivan
Date Added: 07 Sep 2021 13:37
Last Modified: 07 Sep 2021 13:37
Related URLs:
URI: http://irep.ntu.ac.uk/id/eprint/44130

Actions (login required)

Edit View Edit View

Views

Views per month over past year

Downloads

Downloads per month over past year