A hybrid CNN plus BILSTM deep learning-based DSS for efficient prediction of judicial case decisions

A gradual increase in the historical data available in the legal domain over the years has compelled industry experts to collect, assemble, and analyze this data to predict judicial case decisions. However, using this legal data to predict and justify court decisions is no small feat. At present, existing studies on predicting court decisions using limited-size datasets from experimentation have produced a variety of unexpected estimates that were made using machine learning (ML) classifiers with the assistance of common approaches for encoding categorical data. These early works also used convolutional neural networks (CNN), a class of deep neural networks; to extract features without keeping track of sequencing information. This present study proposes predicting court decisions by applying a hybrid deep learning (DL)-based decision support system (DSS); namely CNN with bidirectional long/short-term memory (BiLSTM); to efficiently predict court decisions from historical legal data. Using feature selection, only the most appropriate features were chosen by prioritizing and selecting features that ranked high in the given legal data set. The CNN + BiLSTM hybrid model was then used to predict judicial case decisions. In comparison to other similar studies, the experimental findings of this hybrid model were encouraging; 91.52 % accuracy, 91.74 % precision, 89.04 % recall, and an F1-score of 90.44 %.