Data analysis for USEN

Abstract

Sales predictions is a significant issue for various organizations associated with assembling, co ordinations, advertising, wholesaling and retailing. Modern brands are coordinating new advancements to record and use information to improve their proficiency. A colossal storehouse of Big information is created every day. While mechanical brands are producing exceptionally circulated information from different administrations and applications, a few difficulties in information examination require new ways to deal with help the huge information time. These difficulties for modern enormous information investigation are continuous examination and dynamic from huge heterogeneous information sources in mechanical space. The measures are compulsory to oblige process speed of exchange and to upgrade the normal development in information volume and client conduct. Hence, huge information investigation is an ebb and flow territory of innovative work. The main objective of this project is to use approaches of data analytics, machine learning and processing of historical data for deriving valuable insights from data for USEN through predicting sales. These predictions lead to efficient decision making, enterprise planning and human behavior analysis. The repossessed data from USEN was gone under numerous data mining practices like pre-processing technique. We applied few exploratory analysis procedures on the data also. Several statistical values like p-value, range of the data were analyzed to see which one is the best possible model for the data. ARIMA model was found to be the best suited model as the data was found to be of time series nature. In order to cross validate ARIMA model its results were compared with results with another model. The model which we used for training and cross validation was LSTM, which applied the concept of deep learning and recurrent neural network. The end results were the forecasted values of the data given.