Development of Predictive Analytics Models for Measuring and Improving Business Performances

This is a research project which is focused on application of predictive analytics methods in cheese production plantation. The aim of the project is to use predictive methods in order to predict quantitative sales future outcome in order to plan production of different product categories and also improve business performance in general. The Subject of the research is the retail daily sales volume of different products throughout the period of 14 consequent time units (months). The object is a cheese production factory from the North-West Part of North Macedonia which produces and sales its own cheese products through the wholesale chain and retail supply chain channels. The subject of research is the retail volume of sales of different cheese products. Main goal of the research is to develop a model which can be used in predicting future volume of sales by different products. The research project is practical project for developing predictive analytics model as tool for further business analysis and measurement and improvement of the companies business performance. The research also evaluates and assess academical and practical accumulated know-how of application of predictive analytics cases as an instrument for measurement and improvement of business performance. Available tools, instruments, statistical and other available techniques have also been reviewed, with accent on the forecasting methods and their application for sales forecasting business cases. Predictive analytics is getting on more and more popularity as a comprehensive and reliable analysis for supporting strategic business decisions. The project itself was guided and proceeded according to the CRISP -DM methodology for conducting and developing such predictive analytics model. The main phases of the research project are: Business Understanding, Data Understanding, Data Preparation, Modeling and Evaluation of the models. Many tasks were conducted within the main phases, such as the data gathering. Data was gathered internally from the accounting system of the company and externally, from the national statistical bureau and other free available resources. As a result, 3 models were built and evaluated in details: Time Series with ARIMA Methods, Time Series with Exponential Smoothing and XG Boost Tree. Best results were obtained with the ARIMA method, because this method is mostly suitable for dataset like this with seasonality and fluctuations. Feature importance was also measured for various factors. Results were translated into recommendation for improving the performance of the business as well as recommendations for improving the predictive model itself, as it is not a one time project but rather a circle, striving for continuous improvement.