On some methods of feature engineering useful for craniodental morphometrics of rats, shrews and kangaroos / Aneesha Pillay Balachandran Pillay

Aneesha Pillay , Balachandran Pillay (2024) On some methods of feature engineering useful for craniodental morphometrics of rats, shrews and kangaroos / Aneesha Pillay Balachandran Pillay. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (310Kb)
	PDF (Thesis PhD) Download (2044Kb)

Abstract

This study examines the craniodental morphology of biological organisms using functional data analysis (FDA). Traditional morphometrics (TM) often uses large numbers of morphometric features to study shape variation among biological organisms. However, this can lead to data redundancy, meaning that the features may contain overlapping information. This study proposes using recursive feature elimination (RFE) method to reduce data dimensionality and select the most important attributes based on predictor importance ranking. RFE was applied to the craniodental measurements of Rattus rattus (R.rattus) data to select the best feature subset for both male and female rats. A comparative study based on machine learning algorithms was also conducted by using all features and the RFE-selected features to classify the R. rattus sample based on the age groups. The results showed that the RFE-selected features were able to improve the classification accuracy of the machine learning algorithms. However, the linear measurements used in TM can only detect changes in size and can be insensitive to geometrical transformations. Therefore, GM is used in the subsequent work as it is more sensitive to changes in shapes. Functional data geometric morphometrics (FDGM) for 2D landmark data is introduced and its performance is compared with the classical GM method. FDGM was applied to 2D craniodental landmark data obtained from 90 crania specimens of three shrew species based on three craniodental views (dorsal, jaw, and lateral). The discrete landmarks were converted into continuous curves and represented as linear combinations of basis functions. Principal component analysis (PCA) and linear discriminant analysis (LDA) were then applied to the GM method and FDGM method to observe the classification of the shrew species. The results showed that the FDGM approach produced better results in separating the three clusters of shrew species compared to the GM method. Machine learning approaches were also performed using predicted PC scores obtained from both methods (combination of all three craniodental views and individual views). These analyses favoured FDGM, and the dorsal view of the shrew skull was revealed to give the best representation for distinguishing between the three shrew species. This work also introduces FDGM for 3D landmark coordinate data. FDGM and GM were applied to distinguish dietary categories of kangaroos (fungivores, mixed feeders, browser, and grazer) using landmarks obtained from crania of 41 kangaroo extant species. The results showed that FDGM was able to improve the reconstruction error and distinguish dietary categories of kangaroos better than GM. Simulation studies were conducted to show the general effectiveness of FDGM compared to GM method for both 2D and 3D landmark data. The results obtained from the simulation studies and application to real data showed that FDGM performed better than GM when PCA and LDA were employed. Thus, FDGM provides a powerful and flexible framework for analysing shape variation in geometric morphometrics research.

Actions (For repository staff only : Login required)