Applied Intelligent Decision Making in Machine Learning

Himansu Das, Jitendra Kumar Rout, Suresh Chandra Moharana, Nilanjan Dey, Himansu Das, Jitendra Kumar Rout, Suresh Chandra Moharana, Nilanjan Dey

• Publisher: CRC Press
• Year: 2020
• Print ISBN: 9781032728575
• eBook ISBN: 9781000208580
• Topic: Computer Science
• Subtopic: Programming Games
• Index: Computer Science

The objective of this edited book is to share the outcomes from various research domains to develop efficient, adaptive, and intelligent models to handle the challenges related to decision making. It incorporates the advances in machine intelligent techniques such as data streaming, classification, clustering, pattern matching, feature selection, and deep learning in the decision-making process for several diversified applications such as agriculture, character recognition, landslide susceptibility, recommendation systems, forecasting air quality, healthcare, exchange rate prediction, and image dehazing. It also provides a premier interdisciplinary platform for scientists, researchers, practitioners, and educators to share their thoughts in the context of recent innovations, trends, developments, practical challenges, and advancements in the field of data mining, machine learning, soft computing, and decision science. It also focuses on the usefulness of applied intelligent techniques in the decision-making process in several aspects.

To address these objectives, this edited book includes a dozen chapters contributed by authors from around the globe. The authors attempt to solve these complex problems using several intelligent machine-learning techniques. This allows researchers to understand the mechanism needed to harness the decision-making process using machine-learning techniques for their own respective endeavors.

---

1 Data Stream Mining for Big Data

Chandresh Kumar Maurya

Contents

1.1Introduction

1.2Research Issues in Data Stream Mining

1.3Filtering and Counting in a Data Stream

1.3.1Bloom Filters

1.3.2Counting the Frequency of Items in a Stream

1.3.3Count Unique Items in a Data Stream

1.4Sampling from Data Streams

1.5Concept Drift Detection in Data Streams

1.5.1Causes of Concept Drift

1.5.2Handling Concept Drift

1.5.2.1CUSUM Algorithm

1.5.2.2The Higia Algorithm

1.5.2.3Dynamic Weighted Majority Algorithm

1.6Discussion

References

1.1Introduction

There are two kinds of learning, based on how the data is processed. In batch learning, the data is processed in chunks and often offline. Another type of learning is called online learning, usually performed on streaming data. Another name for online learning is incremental learning. Our focus in this chapter will be incremental learning on streaming data.

A data stream is characterized by certain properties according to Gama (2010):

• Unbounded size:
  • Transient (lasts for only few seconds or minutes);
  • Single-pass over data;
  • Only summaries can be stored;
  • Real-time processing (in-memory).
• Data streams are not static:
  • Incremental/decremental updates;
  • Concept drifts.
• Temporal order may be important.

Traditional algorithms developed for offline processing of the data are not suitable for streaming data due to these issues. There are a few models that are incremental such as K-nearest neighbors (KNN) and naive Bayes. But, still, these models cannot cope with all the issues present in streaming data. For example, if there is a concept drift present in the data, a separate concept-drift detection and alleviation technique needs to be developed. Many researchers have proposed various models for streaming data which tackles some of the issues above. We will look at some of these techniques in detail in the coming sections.

As shown in Figure 1.1, a stream processor takes as input a stream of data such as 0, 1, 1, 0, 1, …. The processor can sample from the stream and process it in main memory, such as when answering a query, and the result of the processing can be dumped into the back-end, such as the hard drive, for downstream tasks if required. Most of the algorithms presented in the subsequent sections will be based on this generic model.

Figure 1.1 General stream processing model.

Data stream mining finds application in several domains such as answering user queries over the web, sensor data (Das et al., 2019; Dey et al., 2019), analyzing network packets, patient health monitoring, and surveillance systems to name a few. Because of the vast application area of the field, I will present some selected case studies in the field at the end of the chapter.

The rest of the chapter is as follows. Re...

Table of contents

• Cover
• Half Title
• Series Page
• Title Page
• Copyright Page
• Table of Contents
• Preface
• Notes on the Editors and Contributors
• 1 Data Stream Mining for Big Data
• 2 Decoding Common Machine Learning Methods Agricultural Application Case Studies Using Open Source Software
• 3 A Multi-Stage Hybrid Model for Odia Compound Character Recognition
• 4 Development of Hybrid Computational Approaches for Landslide Susceptibility Mapping Using Remotely Sensed Data in East Sikkim, India
• 5 Domain-Specific Journal Recommendation Using a Feed Forward Neural Network
• 6 Forecasting Air Quality in India through an Ensemble Clustering Technique
• 7 An Intelligence-Based Health Biomarker Identification System Using Microarray Analysis
• 8 Extraction of Medical Entities Using a Matrix-Based Pattern-Matching Method
• 9 Supporting Environmental Decision Making Application of Machine Learning Techniques to Australia’s Emissions
• 10 Prediction Analysis of Exchange Rate Forecasting Using Deep Learning-Based Neural Network Models
• 11 Optimal Selection of Features Using Teaching-Learning-Based Optimization Algorithm for Classification
• 12 An Enhanced Image Dehazing Procedure Using CLAHE and a Guided Filter
• Index