SVM Classification Using Segmented RBF Kernels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification systems face challenges in efficiently analyzing and classifying large data sets, particularly with Support Vector Machines (SVMs) due to non-linear separability and increasing training time with larger data sets, necessitating a more effective method for user classification.
Innovation Solution
A computer-implemented classification system that segments data sets using decision trees, identifies difficult regions, computes standard deviations, selects radial basis function (RBF) kernels, and generates SVM models to create hyper planes for user categorization, leveraging a Map Reduce framework for distributed processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Support Vector Machines are used for classification, then classification accuracy is improved, but training time increases dramatically with larger data sets
Solution Approach 1:
The patent divides the large data set into multiple segments or partitions, allowing the SVM training process to be distributed across multiple computing nodes. Each node trains on a subset of the data independently, significantly reducing the training time for each individual model while maintaining overall classification accuracy through aggregation of results.
2Measurement precision
If a suitable Kernel is chosen to map to linearly separable feature space, then classification capability is improved, but system complexity increases
Solution Approach 1:
The patent applies different kernel functions to different segments of the data set based on their local characteristics. Instead of using a single global kernel for the entire data set, the system selects and applies appropriate kernels locally to each segment, improving classification capability for non-linearly separable regions while keeping the overall system manageable through localized complexity.
Data Source
AI summary
Computer implemented system and method for classifying users is provided. The method includes obtaining a data set from one or more sources corresponding to one or more users; segmenting the data set into segments, each segment comprising one or more data points, wherein each data point is specific and unique to a user; identifying difficult regions from one or more segments; computing a standard deviation of distances between the data points within each of the identified difficult regions for each of the identified difficult regions; selecting a radial basis function (RBF) kernel based on the standard deviation computed for each of the identified difficult regions; generating a support vector machine model on each of the identified difficult regions based on the selected RBF kernel to obtain a hyper plane comprising support vectors and weight assigned to the vectors; and classifying the users into a category based on the hyper plane.