Sales prediction using data analytics: A machine learning approach

Artificial intelligence applications are attracting increasing amounts of attention from researchers. These applications, which are mostly used in the fields of robotics and informatics, are able to solve problems experienced on a daily basis in industry. Machine learning (ML), a sub-branch of artificial intelligence, enables predictions to be made about the future based on historical data, and it is therefore possible to use ML methods to make sales forecasts. Since the amounts of data stored by companies are now very large, sales forecast operations are now fairly difficult with traditional methods. In this thesis, a sales forecast is made for a company in the office supply industry using ML algorithms. First, statistical and visual inferences were made from the dataset using exploratory data analysis. The dataset was examined by considering categorical variables and numerical variables separately, and missing data and outlier analyses were carried out. Twelve algorithms (Linear regression, Ridge regression, Lasso regression, ElasticNet regression, K-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Regression (SVR), Classification and Regression Trees (CART), Gradient Boosting Machine (GBM), Light Gradient Boosting Machine (LightGBM), Categorical Boosting Machine (CatBoost) and Extreme Gradient Boosting Machine (XGBoost)) were applied, and the error functions were evaluated using the Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and coefficient of determination (R^2). Model tuning was performed using different parameters for each algorithm. The results showed that the most successful algorithm was XGBoost, with a value for RMSE of 11.7643. A prediction was made based on this tuned score using XGBoost algorithm model.