In this section, we present our methodology followed in conducting a systematic literature review, and then we will discuss the results found in the related works.
Literature review searching methodology
The title Moon phase observation using machine learning is not frequent. So, we have conducted our search for all types of papers and for all years without exception. We have searched several digital resources and databases, such as IEEE Xplore, Science Direct, Scopus, Springer Link, and Google Scholar, for related articles. Several keywords were used for the searching process such as (‘the’ and ‘month’ and ‘of’ and ‘Ramadan’), (‘Hijri’ and ‘calendar’), (‘Moon’ and ‘phases’), (‘machine’ and ‘learning’), (‘Moon’ or ‘observation’), (‘lunar’ or ‘crescent’), (‘lunar’ or ‘visibility’), (‘crescent’ or ‘visibility’), and (‘machine’ and ‘learning’ and ‘for’ and ‘Moon birth’ or ‘visibility’).
Figure 5 shows the process followed in selecting the related articles found in our search. A total of 133,509 articles were identified. Then, a screening process was conducted to exclude articles not really relevant to our study, and we have selected only articles which are related to computer science, astronomy, machine learning, artificial intelligence, and data analysis. We ended up with 14,898 articles, which have been passed on a second screening process after screening topics and subjects. After careful reading of these articles, and therefore, only 46 papers have been identified and summarized in the related work sub-section.
Literature review searching process.
Related work
As stated earlier, many societies depend on the lunar phase cycle for their calendar The establishment of the lunar calendar is based on the observation of the first sighting of the new crescent Moon. Most Islamic countries and Muslims across the world depend on the Islamic calendar converter, which is based on the arithmetical (or tabular) calendar to predict the approximate beginning of the new month. This arithmetical calendar was introduced earlier by Muslim astronomers during the ninth century. However, this predicated calendar has an error of 4 × 10–4 days per 30 years, which accumulates to a whole day in 2492 years15.
It is obvious that in order for the new crescent Moon to be sighted, a minimum brightness contrast between the Moon and the sky is required. Starting with the criteria set by the Babylonians, who stated that the visibility of the new crescent Moon with naked-eyes at local Sunset was subject to two conditions: the age of the Moon must be more than 24 h and the Moon’s lag time must be greater than 24 min11. Al-Khwarizmi’s criterion for the new crescent Moon to be sighted was the arc of separation of the Sun and Moon along the celestial equator must be greater than 12°. On the other hand, Danjon’s criterion16 for the new crescent Moon visibility with the naked-eyes was that the ARCL must be at least of 7°. Al-Battani17 suggested the use of the parameters DAZ, Moon altitude, and the width of the Moon, W, for the visibility of the new crescent Moon. Ilyas18 set the condition for the new crescent Moon to be sighted is that the ARCL must be at least in the range of 9° to 10°. McNally19 argued Danjon’s work stating that the new crescent Moon shortening is mainly due to atmospheric conditions. In another study20, Ilyas updated the condition of the new crescent Moon visibility for the ARCV to be 10.5°. According to the authors in21, the visibility of the new crescent Moon is unpredictable using only one of the above-mentioned parameters. For example, with only the Moon Age or Lag, there is no predictive value. In order to correctly calculate the Moon’s appearance, several astronomical conditions are combined to calculate the Moon’s appearance, such as the minimum altitude of the new crescent Moon above the horizon, the minimum ARCL, and the minimum age of the Moon since the conjunction occurrence.
Doggett et al.22 reported the new crescent Moon observation results of five monophases throughout North America. They found that ancient and medieval observation rules are highly unreliable. They also claimed that recent empirical criteria such as the relative altitude and azimuth of the Moon at Sunset time would reflect a very good observation accuracy. They claimed that information about atmospheric, optical, and human factors would have an effect on observations. Yallop23 examined Maunder9, Indian Astronomical Ephemeris and Bruin10 methods. The first two methods considered the parameters DAZ and ARCV, while the last method considered ARCV as a function of Yallop came up with the formulae of the time of the best visibility of the new crescent Moon. Also, Schaefer24 and based on over 200 observations, suggested that the atmospheric conditions must be considered in the new crescent Moon visibility process. He argued that using the empirical method cannot be applied in all situations, and this is because it depends on the observer’s location and other factors. For that, Schaefer developed an algorithm to predict the visibility of the new crescent Moon. The program takes as input the set of observing conditions such as the date of the new Moon, observer location in terms of latitude and longitude, observer age, atmospheric conditions, and the program generates the age of the Moon at the time of the first visibility for the given conditions. Fatoohi et al.25, on the other hand, discussed all the work suggested by Danjon, Illyas, McNally, and Schaefer and concluded that the minimum value of the ARCL, which is 7.5°, is a reasonable value for the new crescent Moon to be visible.
In 1998, the Islamic Crescent Observation Project (ICOP for short), an initiative of the Arab union of astronomers and space sciences and the Jordanian astronomy society11, was established to serve as a central point for gathering information about the observation of the new crescent Moon at the beginning of each month and this from different countries across the world and creating a central database for that. Odeh11 claimed that the accurate prediction of the visibility of the crescent Moon could not be achieved by a single parameter. Furthermore, he added that the Moon’s age is not a good indicator of its brightness. Odeh proposed a new criterion for the prediction of the visibility of the crescent Moon, which is the W (in arc minutes), and which must be used in conjunction with the ARCV, yielding the equation of visibility prediction of the new crescent Moon, Odeh suggested four possible scenarios based on the value of V to decide whether the new crescent Moon is visible with naked-eyes, by optical aid and can be seen by naked-eyes, visible only by optical aid, not visible at all. The other important Odeh contribution was the creation of a database of lunar crescent observations from different sources as part of the ICOP project to come up with a total of 737 observations. Caldwell26 studied the effects of the Moon’s lag time and the ARCL on the visibility of the new crescent Moon with naked-eyes or binoculars. He conducted a simulation and derived equation setting boundaries where visibility criterion lines were suggested. He derived the values of the Moon’s lag time as a function of the ARCL to find out the three lines to confirm the visibility of the new crescent Moon, whether with naked-eyes or binoculars or the new crescent Moon visibility is impossible. Özlem’s work27 aimed to propose a visibility criterion, which can also be used for thicker crescents, including daytime visibility, when the Moon is visible together with the Sun.
Alrefay et al.28 claimed that predicting the visibility of the new crescent Moon is a challenging and difficult process for several technical reasons. The lunar calendar cannot be calculated on the basis of observations since the start of the next month requires about 29 days of waiting. Such observation calendars are highly influenced by atmospheric conditions and usually involve errors caused by human factors such as illusions. The authors in this study applied several criteria from 545 observations over 27 years (1988–2015) from different locations in Saudi Arabia. They developed a new criterion based on the lunar W and the ARCV. According to the 1978 Turkish International Conference, the majority of Muslim countries agreed that the Islamic month begins when the ARCL is greater than 8° and the new Moon height from the horizon when the Sunset must be greater than 5°29,30.
Some researchers suggested systems or applications to help in the detection of the first crescent Moon or the computation of the Hijri calendar. For instance, Alhammadi et al.31 proposed an electronic system that uses inputs from a database; the inputs suggest the location to which the system camera should be oriented. This location is based on the DAZ and the Moon’s latitude. Then, the images captured by the camera are processed to confirm whether these images correspond to new crescent Moon or not. A desktop application was proposed in32 to compute Hijri calendar dates and displays all Islamic events in the year.
Recently, the advent of computational methods and the availability of the immense data generated from satellites in orbit spaces have changed the way of analyzing and processing the data. For example, NASA’s Space Program provides us with massive data on cosmic objects to help us exploring space. Astronomy experts have examined images generated and collected by telescopes like Hubble. Later, with the introduction of advanced telescope technologies and satellites such as Kepler, these have opened new horizons to autorotate the analysis process of observation and the need to learn the computer machine to do the tasks. These satellites and telescopes process data using various features and imaging techniques, not only to capture images but also to identify these exoplanets33,34. In their paper, Moshayedi et al.35 have presented a prototype system that utilizes machine learning models to detect the various phases of the moon. The primary focus of their research is centered on analyzing a collection of moon images to determine the lunar phase that corresponds to each image. It is worth noting that their approach differs from our own work, as our investigation is based on a dataset consisting of numerical data that pertains to distinct moon phases, rather than image analysis. Our specific aim was to identify the appearance of the new crescent moon, which serves as the marker for the commencement of the Hijri month. The authors in36 proposed an image processing-based method to detect the crescent Moon from observed images. The processing went through several phases like a Gaussian smoothing filter to remove noise, followed by the quality enhancement of the image focusing on the part containing the crescent Moon. The authors used the circular Hough transform to extract the crescent Moon from the background of an image and detect the center of the Moon and the shape of the crescent. In the same context, Sejzei and Jamzad37 developed a toolbox in Matlab that enhances the image of a crescent Moon and helps observers to see/detect the crescent Moon.
Artificial intelligent techniques and algorithms such as Machine (ML) and Deep Learning (DL) algorithms represent powerful tools in processing the immense data of extrasolar planets. More recently, researchers have suggested the use of ML algorithms for the classification of astronomical objects38,39,40,41,42,43. Some of these common algorithms which were applied to classify objects found in the Kepler Cumulative Object of Interest are Random Forest (RF), Support Vector Machines (SVM), AdaBoost, and Deep Neural Networks (DNN)33,41. Beniwal et al.40 applied classical ML algorithms for pulsar classification.
Tafseer44 considered the problem of Moon crescent visibility for each month as a classification problem and suggested the use of ML algorithms instead of mathematical or astronomical methods. His study was based on the dataset found on the website managed by ICOP. A pre-process of the dataset was conducted where 1070 samples with cloudy or partially cloudy sky condition were removed. The author added some astronomical features such as the age of the Moon, the Moon’s lag time, the altitude difference, DAZ, the Moon phase, and the atmosphere. Four algorithms were considered in this study, namely logistic regression, Neural Network, SVM, and RF, applied to 1522 samples, among which 80% were used for the training phase and 20% were used for the testing phase. Results obtained in this study showed that RF achieved better precision in predicting the visibility of the crescent Moon with 88%. In another study45, the author employed an Artificial Neural Network (ANN) model to forecast the visibility of the new crescent moon. However, the research was limited to the use of a single machine learning algorithm (ANN) and was restricted to a singular geographical location, specifically Iraq.
The unification of Islamic Hijri calendar, and the accuracy of the beginning of each Hijri month, has always been a concern for the Muslim community all over the world. The authors in12,46 raised the discrepancies with the beginning of the holly months in different parts of the world. Khan46 claimed that the only workable solutions is using astronomical calculations to reduce errors. Zainon et al. in12, on the other hand, discussed the criteria and factors to take into consideration in the determination of the beginning of each Islamic Hijri month, such as the conjunction, and the new crescent Moon visibility. They also suggested some required parameters for the Moon to be possibly sighted by the naked-eyes or using a sophisticated tool like telescopes, which are Moon age, Moon’s lag time, Moon’s altitude, ARCL, ARCV, DAZ, and W. These parameters12,28 are presented in Fig. 3.
In an effort to help in this matter, this current study suggests the application of ML algorithms in the prediction of the visibility of the new crescent Moon. To the best of our knowledge, in the context of this research, only44 and45 tackled the prediction of the crescent Moon visibility applying some ML algorithms. Nevertheless, Allawi45 was confined to a single ML algorithm, and it was just restricted to one individual country, Iraq. Referring to the dataset used in44, several limitations were identified such as DAZ feature which should be positive numbers, the values of some features like ARCL (elongation) were not accurate, names of some countries and cities features were not consistent; same countries and cities in different languages or abbreviated, which may affect the analysis results. To overcome all these shortcomings, we built our own dataset and included all the years up to the current year 2022, as explained in “The proposed model” section.
