Project ID
Name:
Level:
Who:
Novelty:
 Classification of Species Using NN
Moderate to high
Students and hobbyists of neural networks
Original

Classification of Species in Kingdom Animalia into Groups Using Neural Networks - A Draft Proposal

This proposal outlines the project subject area, the problem, and my approach to a solution.

Introduction:

Classification of living things is essential for understanding the surrounding world, and classification can be efficiently implemented using neural networks. Classifying any given organism as to its groups is beyond our scope; however, classification of a random organism based on its accurate description into a well defined high order group should be feasible. The design can be useful for children to even mature students as a learning tool, and a refined version may be useful for biologists and database creators.

Problem Definition:

Biologists classify life from general to specific as follows: KingdomàDivisionàClassàOrderàFamilyàGenusàSpecies. The Kingdom is further divided into 5-kingdom system, and Kingdom Animalia consists of common species familiar to us. The Kingdom Animalia consists of following main groups: sponges, cnidarians, flatworms, roundworms, molluses, segmented worms, echinoderms, chordates, amphibians, reptiles, birds, and mammals. Thus, the problem can be stated as follows: given an organism and a set of characterization classify a particular organism into one of above groups. Although, the problem is similar to that of distinguishing between a apple and a banana. the scope, and the relevance of the classification, and the detailed biological knowledge need for the problem makes it much harder.

Methodology:

A casual search on the "google" provides examples of similar previous works. Neural networks has been used to identify microbial life forms, mushrooms, and other life forms. Most of them pictorize the species under consideration, and perhaps such application of neural networks is the most appropriate. However, I want to approach the design in terms of description based classification. (Extracting description based on a picture is a challenging neural network problem itself.) The main task is to classify well defined characteristics of the groups and then to feed that information into a description matrix. Use examples to train the network, and stop when a expected performance level has been achieved.

Preliminary Concerns:

Why not just a slide rule to classify species, what is the specific advantage of using neural networks?
It is automatic, computerized, and its training capability allows for robustness not found in table driven classification.

Have not people already done such things?
Not in the particular manner and for the particular purpose I propose.

Is the project broad, non-technical, or simple for the course?
In my opinion, No. The project requires an implementation of neural network for a real world application, and thats challenging.