PaperlayerView exact revision on arXiv

Source-linked AI summary

Analysis of Heart Diseases Dataset using Neural Network Approach

K. Usha Rani

arXiv:1110.2626v1cs.LGcs.DB

TL;DR

The paper addresses classification of medical data, specifically heart-disease cases. It trains neural networks with back-propagation and parallel processing, and reports satisfactory classification results.

  • Problem

    The paper addresses classification of medical data, including distinguishing normal from abnormal cases in a heart-disease dataset.

    Source: §1 · p. 1, §4 · p. 3

  • Method

    The experiment trains feedforward neural networks with back-propagation, momentum, and variable learning rate, with parallel processing across neurons in hidden and output layers.

    Source: §4.1 · p. 5, §5 · p. 6

  • Results

    The authors report that neural-network techniques provide satisfactory results for the heart-disease classification task.

    Source: §5 · p. 7

  • Takeaways & Limitations

    The experiment assigns cases to four classes: normal, first stroke, second stroke, and end of life.

    Source: §4.1 · p. 5

Abstract

from arXiv · show

One of the important techniques of Data mining is Classification. Many real world problems in various fields such as business, science, industry and medicine can be solved by using classification approach. Neural Networks have emerged as an important tool for classification. The advantages of Neural Networks helps for efficient classification of given data. In this study a Heart diseases dataset is analyzed using Neural Network approach. To increase the efficiency of the classification process parallel approach is also adopted in the training phase.

1. INTRODUCTION

Data mining seeks patterns and predictions from data, with classification building a model from labeled examples. This study applies a neural-network classifier to medical data and uses parallel processing in the network.

1. INTRODUCTION

  • Data mining supports knowledge discovery by uncovering relationships in data and predicting outcomes.The paper describes data mining as part of the broader Knowledge Discovery in Databases process.Source: §1 · p. 1
  • The introduction places the study within data mining’s applications across business, science, industry, and medicine.Source: §1 · p. 1
  • Classification uses a training set with attributes and class labels to build a model that assigns labels to new data.Source: §1 · p. 1
  • Neural networks are presented as one soft-computing method for extracting patterns, alongside fuzzy logic, genetic algorithms, and rough sets.Source: §1 · p. 1
  • The study proposes a neural-network approach trained with back-propagation and analyzes its performance on different test data.Parallel processing is implemented at neurons in different network layers.Source: §1 · p. 2

2. Neural Networks

Neural networks comprise interconnected processing elements and learn by adjusting connection weights. The section describes their training, common back-propagation architecture, and potential for parallel implementation.

2. Neural Networks

  • A neural network comprises processing elements connected by weighted links, and training adjusts the weights to produce desired outputs.Source: §2 · p. 2
  • Neural networks support parallel implementation at the layer level and are described as suitable for noisy or uncertain information.The section also identifies fault tolerance as a characteristic of ANN.Source: §2 · p. 2
  • Networks learn iteratively by presenting data cases one at a time and adjusting input weights to predict their class labels.Training starts with randomly chosen initial weights.Source: §2 · p. 2
  • The section focuses on feedforward multilayer networks trained with back-propagation, a widely studied neural-network classifier.The typical architecture has input and output layers with at least one hidden layer.Source: §2 · pp. 2–3
  • Back-propagation computes outputs in a forward sweep, compares them with desired outputs, then propagates the difference backward to adjust weights.Source: §2 · p. 3

3. NEURAL NETWORKS IN THE MEDICAL FIELD

The paper situates neural networks within medical applications and motivates their use for disease classification. It describes extracting rules from disease classifications to predict disease stages.

3. NEURAL NETWORKS IN THE MEDICAL FIELD

  • Neural networks have been applied to medical tasks including diagnostic support, biochemical analysis, image analysis, and drug development.Examples include tumor detection in ultrasound images, mammogram microcalcification classification, chest x-ray classification, and MRI tissue and vessel classification.Source: §3 · p. 3
  • Neural networks can extract rules from disease classifications, which can be used to predict whether someone will have a particular disease stage.Source: §3 · p. 3
  • The section motivates data mining in healthcare as a way to find patterns in growing stores of data for patient care and diagnostic capabilities.Source: §3 · p. 3

4. EXPERIMENT - CLASSIFICATION OF HEART DISEASES DATASET

The experiment uses Cleveland heart-disease data to train single- and multilayer neural networks, then assesses classification by comparing test outputs with target outputs.

4. EXPERIMENT - CLASSIFICATION OF HEART DISEASES DATASET

  • The Cleveland dataset contains 414 instances and is described as classifying people as normal or abnormal, with 13 attributes and a class attribute.Source: Data Representation · p. 3, §4 · p. 3

4.1 Training the Neural Network

  • Feed-forward networks with and without hidden layers use backpropagation with momentum and variable learning rate; 13 input neurons represent 13 attributes, while two output neurons represent four classes.Source: §4.1 · p. 5

4.2 Performance of the Network

  • Table 1 presents experimental results; performance is assessed by comparing trained-network outputs on test samples with target outputs.Source: Table 1 · p. 6, §4.2 · p. 6

5 CONCLUSION

The paper applies single- and multilayer neural networks to heart-disease classification, with parallelism at hidden and output neurons intended to speed learning. It reports satisfactory classification results.

  • The study evaluates single- and multilayer neural networks trained with backpropagation, momentum, and a variable learning rate on heart-disease data.Source: §5 · p. 6
  • The experimental results are reported to provide satisfactory results for the heart-disease classification task.Source: §5 · p. 7
  • Parallelism at each neuron in hidden and output layers is implemented to speed the learning process.Source: §5 · p. 7
Loading 1110.2626v1…