Classification of Grievances in Hello Sarkar using Supervised Machine Learning

Abstract

Classification of Grievances automatically as per some pre-defined category is an automated task that can be carried out with the aid of some confidence gained after learning the content from a training set of grievance dataset. This research mainly aims to evaluate widely used supervised machine learning approaches like Naïve Bayes and SVM. For this experiment, the dataset of 4 categories with about 3,000 datasets collected from the official site of ‘Hello Sarkar’ portal which has been used by the Government of Nepal, Office of the Prime Minister and Council of Ministers for handling the public grievances was used. As per the recent research outcomes, none of the researchers had spoken about which machine learning algorithm outperforms the rest of the other and has always suggested that the outcome solely relies on the type of dataset used. So, in this research, above -mentioned machine learning was used and the accuracy of those applied algorithms was compared. As public grievances are either in Nepali, English or in Roman form, the research procedure mainly focused on how well the above-mentioned algorithms perform on those grievances in Nepali language dataset. Moreover, as per the suggestions from various researches conducted previously, the enhancement in accuracy was targeted out by cleaning the dataset by adding the stop words in NLTK corpus and removing extra useless symbols during preprocessing state for refining dataset before the implementation of machine training algorithms.