Feature Selection Method Based On Statistics of Compound Words for Arabic Text Classification

One of the main problems of text classification is the high dimensionality of the feature space. Feature selection methods are normally used to reduce the dimensionality of datasets to improve the performance of the classification, or to reduce the processing time, or both. To improve the performance of text classification, a feature selection algorithm is presented, based on terminology extracted from the statistics of compound words, to reduce the high dimensionality of the feature space. The proposed method is evaluated as a standalone method and in combination with other feature selection methods (two-stage method). The performance of the proposed algorithm is compared to the performance of six well-known feature selection methods including Information Gain, Chi-Square, Gini Index, Support Vector Machine-Based, Principal Components Analysis and Symmetric Uncertainty. A wide range of comparative experiments were conducted on three Arabic standard datasets and with three classification algorithms. The experimental results clearly show the superiority of the proposed method in both cases as a standalone or in a two-stage scenario. The results show that the proposed method behaves better than traditional approaches in terms of classification accuracy with a 6-10% gain in the macro-average, F1.