Generating Bengali Explanations for Machine Learning Predictions

The popularity of machine learning (ML) models is increasing day by day, especially for prediction tasks. Consequently, people are using ML models to make predictions for different tasks in Academia as well as in industry. These predictions help people to make better decisions such as the decision of investing in a particular share can be made after predicting the price of that share. However, while making these decisions, people often get confused about the predictions made by ML models and they want to know why the ML models are making those predictions. Hence, it is important to explain the predictions of the ML models. There is a research that can explain the predictions for image prediction tasks using relation information of the objects that are present in the predicted image. That research learns rules using the extracted information of the predicted images and using the learned rules the explanations are generated in English language. In that research, the relations of the objects are identified manually while learning the rules for generating explanations. In our research, we want to apply ML techniques to extract information of the images including the relation information of the objects presented in images. We also want to explore that how the extracted information can be used to generate explanation in Bengali. In our work, we will explore if it is possible to generate the explanation using the extracted image information without learning the rules. Our research will improve the explanation generation process using relation information by applying ML technique to identify the relations of objects in images and will generate explanations in Bengali that will help the Bengali people to understand the predictions of ML models.