INDONESIA PURCHASING POWER PREDICTION USING REGRESSION ANALYSIS

• Main Author: Muhammad Dyasputro, Drestanto
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/49893
• Institution: Institut Teknologi Bandung

Purchasing power is the ability of the community as consumers to buy goods or services needed. Purchasing power must be known to analyze the economic behavoiur of a country. Currently, quarterly data of purchasing power can be calculated at the end of the quarter. However, the value purchasing power is also needed at any time in a quarter. This study will use supporting data in the form of price index data and changes in consumption of goods to predict the Indonesia's purchasing power. The data characteristics used are numerical data which are time series with a small number of datasets (only about 20-30 instances). Prediction is done using regression techniques. The methodology used in this research is CRISP-DM (Cross Industry Standard Process for Data Mining) methodology. The methodology stages include data understanding, data preparation, modelling, and evaluation. There are 20 features used in this study, consisting of 9 food price features and 11 non-food features. With so many features, the modelling stage utilizes feature selection techniques to improve model performance. Important features used in this research are car sales, world oil prices, syariah loans, and syariah third party funds. In building the model, other techniques such as stationarity test, differencing, z-score calculation, and cross validation evaluation are also used. Learning techniques used in this study are LSTM (long short term memory), SVR (support vector regression), and Random Forest. In this study, three candidate models (machine learning models) were used that were compared to their RMSE (root mean squared error). The results showed that the best model was SVR, because it has the smallest error (RMSE = 0.2881). Compared to other models, they are Random Forest (RMSE = 0.4407) and LSTM (RMSE = 0.8531). Quarterly test results produce RMSE = 0.2624 and monthly testing results produce RMSE = 0.4589.