本文为印度Rourkela国立技术研究所（作者：Priyanka Mehtani）的学士论文，共45页。

分类是一种数据挖掘（机器学习）技术，用于预测数据实例的群组成员关系。模式分类包括构建一个函数，将输入特征空间映射到两个或两个以上类的输出空间。神经网络是模式分类领域的一种有效工具，利用训练和测试数据建立模型。然而，神经网络的成功在很大程度上取决于训练过程的性能，即训练算法。为了提高神经网络的性能，目前已经提出了许多训练算法。

在本课题中，我们将对前馈神经网络的三种训练算法——反向传播算法、改进的反向传播算法和光学反向传播算法进行比较研究，这些算法仅根据其误差函数不同而有所不同。我们将使用这些算法对神经网络进行训练，并从IRIS数据集（取自UCI库，然后归一化）中获取75个实例，每类25个。达到准确度所需的周期总数称为收敛速度。比较过程的基本准则是收敛速度和分类精度。为了检验这三种训练算法的效率，绘制了周期数与均方误差（MSE）之间的关系图，训练过程持续到MSE降至0.01，同时观察了动量和学习速率对算法性能的影响，然后扩展比较了多层前馈网络与概率网络的性能。

Classification is a data mining (machinelearning) technique used to predict group membership for data instances.Pattern Classification involves building a function that maps the input featurespace to an output space of two or more than two classes. Neural Networks (NN)are an effective tool in the field of pattern classification, using trainingand testing data to build a model. However, the success of the networks ishighly dependent on the performance of the training process and hence thetraining algorithm. Many training algorithms have been proposed so far toimprove the performance of neural networks. In this project, we shall make acomparative study of training feedforward neural network using the threealgorithms - Backpropagation Algorithm, Modified Backpropagation Algorithm andOptical Backpropagation Algorithm. These algorithms differ only on the basis oftheir error functions. We shall train the neural networks using thesealgorithms and taking 75 instances from the iris dataset (taken from the UCI repositoryand then normalised) ; 25 from each class. The total number of epochs requiredto reach the degree of accuracy is referred to as the convergence rate. Thebasic criteria of comparison process are the convergence rate and theclassification accuracy. To check the efficiency of the three trainingalgorithms, graphs are plotted between No. of Epochs vs. Mean Square Error(MSE). The training process continues till M.S.E falls to a value 0.01. Theeffect of using the momentum and learning rate on the performance of algorithmare also observed. The comparison is then extended to compare the performanceof multilayer feedforward network with Probabilistic network.

1 引言

2 文献回顾

3 相关概念

3.1 数据挖掘

3.2 人工神经网络

3.3 基于前馈神经网络的分类

3.4 基于概率神经网络的分类

4 仿真结果

5 讨论

6 结论

更多精彩文章请关注公众号：