A Robust Deep Model for Improved Categorization of Legal Documents for Predictive Analytics

Abstract

Predictive legal analytics is a technology used to predict the chances of successful and unsuccessful outcomes in a particular case. Predictive legal analytics is performed through automated document classification for facilitating legal experts in their classification of court documents to retrieve and understand the details of specific legal factors from legal judgments for accurate document analysis. However, extracting these factors from legal texts document is a time-consuming process. In order to facilitate the task of classifying documents, a robust method namely Distributed Stochastic Keyword Extraction based Ensemble Theil-Sen Regressive Deep Belief Reweight Boost Classification (DSKE-TRDBRBC) is proposed. The DSKE-TRDBRBC technique consists of two major processes namely Keyword Extraction and Classification. At first, the t-distributed stochastic neighbor embedding technique is applied to DSKE-TRDBRBC for keyword extraction. This in turn minimizes the time consumption for document classification. After that, the Ensemble Theil-Sen Regressive Deep Belief Reweight Boosting technique is applied for document classification. The Ensemble boosting algorithm initially constructs’ set of Theil-Sen Regressive Deep Belief neural networks to classify the input legal documents. Then the results of the Deep Belief neural network are combined to built a strong classifier by reducing the error. This aids in improving the classification accuracy. The proposed method is experimentally evaluated with various metrics such as F-measure , recall, accuracy, precision, , and computational time. The experimental results quantitatively confirm that the proposed DSKE-TRDBRBC technique achieves better accuracy with lowest computation time as compared to the conventional approaches.