Abstract: Content quality is an important factor for journal impacts.Sci-tech journals choose the forefront of academic study as their publishing topic,and the perspectiveness and innovation of topics will be crucial for journals to increase influence. At present most journals determine their topics through expert recommendation or simple statistical analysis on specific objective in specific field,which lacks the universal and feasible topic planning and prediction model.A topic planning method of sci-tech journals in electric engineering field based on academic popularity was proposed.The deep forest model was used to analyze the popularity data of topic words and influential data of papers.The prediction model consists of multi-grained feature scanning and cascading ensemble learning,which provides advice of choosing topics for sci-tech journals in electric engineering field.Experimental results shows that the proposed topic planning method based on deep forest has an acceptable training time and its average value of prediction errors is the lowest compared with deep neural network and random forest algorithm.