2018英特尔ISEF获奖作品摘要 系统软件 Intel Finalist Abstract Systems Software

获奖作品基本信息

年份	2018
学科	系统软件 Systems Software
国家/州	United State of America

获奖作品名称

A Novel Machine Learning Approach for Determining the Confounding Factors for Cancer Identification: An Integration of Neural Learning and Decision Tree

获奖作品摘要

Cancer is a heterogeneous disease comprising different subtypes. Its occurrence alters clinical parameters in human body. The exact cause being unknown, various internal factors, demographic and lifestyle related issues are found responsible. Identification of a set of confounding factors of cancer can be of immense help for early diagnosis.

In this project, we try to automatically identify the major factors causing cancer using machine learning. However, paucity of cancerous samples is the chief bottleneck. We design a novel neural network and decision tree based machine learning algorithm that can identify the set of main causal factors of cancer yielding more than 95% detection accuracy. This helps predict occurrence of cancer in patients.

Initially a generative restricted Boltzmann Machine (RBM), a supervised deep neural network, is used to model the bigger (non-cancerous) class. Functioning as one-class classifier, it confidently determines the non-cancerous samples (using the principle of free energy) to be removed from the dataset.  We also devise a heuristic based on inter-intra class distance ratio to prune noisy samples. The resultant near-balanced set is used to find out the causal factors using decision tree. The built-in tree is later used for cancer detection.  A multilayer perceptron based similar system is also designed.

Experiments are carried out with some benchmark datasets. The system is compared with existing Tomek link based pruning and 1-NN classifier for handling imbalanced data. An average increase in accuracy of 5%-50% is noticed. Moreover, the identified factors corroborate to medical literature. To further elicit the robustness of the system, experiments are conducted on datasets of other diseases and the enhancement is 5%-25%.

---

高中生科研 英特尔 Intel ISEF
资讯 · 课程 · 全程指导
请扫码添加微信好友

---

高中生科研学术活动 英特尔 Intel ISEF 简介

英特尔国际科学与工程大奖赛，简称 "ISEF"，由美国 Society for Science and the Public（科学和公共服务协会）主办，英特尔公司冠名赞助，是全球规模最大、等级最高的中学生的科研科创赛事。ISEF 的学术活动学科包括了所有数学、自然科学、工程的全部领域和部分社会科学。ISEF 素有全球青少年科学学术活动的“世界杯”之美誉，旨在鼓励学生团队协作，开拓创新，长期专一深入地研究自己感兴趣的课题。

>>> 实用链接汇总 <<<

英特尔 ISEF 学术活动详细介绍

英特尔 ISEF 全程指导方案

· 数学 · 物理 · 化学 · 生物 · 计算机 · 工程 ·

学科简介：系统软件 Systems Software

The study or development of software, information processes or methodologies to demonstrate, analyze, or control a process/solution.

Subcategories:

Algorithms (ALG): The study or creation of algorithms - step-by-step procedure of calculations to complete a specific task in data processing, automated reasoning and computing.

Cybersecurity (CYB): Studies involving the protection of a computer or computer system against unauthorized access or attacks. This can include studies involving hardware, network, software, host or multimedia security.

Databases (DAT): Studies that create or analyze data organization for ease of access, management and update.

Human/Machine Interface (HMC): Software application that presents information to a user about the state of a process and to accept and implement the operator’s control instructions.

Languages and Operating Systems (LNG): Studies that involve the development or analysis of artificial languages used to write instructions that can be translated into machine language and then executed by a computer or system software responsible for the direct control and management of hardware and basic system operations of a computer or mobile device.

Mobile Apps (APP): A study involving a software application developed specifically for use on small, wireless computing devices.  These studies may include front-end development techniques, such as user interface design and cross-platform support, and/or back-end development techniques, such as data services and business logic.

Online Learning (LRN): Studies that focus on utilizing electronic technologies to access educational curriculum outside of a traditional means.  Studies explore the design of learning activities and programs with online technologies, as well as the effective use of e-learning systems.

Other (OTH): Studies that cannot be assigned to one of the above subcategories. If the project involves multiple subcategories, the principal subcategory should be chosen instead of Other.