Sheth Preet Ankur

CM4050

Construction industry is challenged with ripe for disruption. Taking scenario of growing infrastructure in India, steel demand is significantly high. Steel bars also called reinforcement bars are one of the most widely used and one of the most expensive construction material. Thermo-Mechanically Treated bars, also known as TMT bars, are high-strength reinforcement bars . The main objective of the study was to forecast the price of steel- particularly reinforcement which will be helpful to many stakeholders of the construction project. Machine learning was used to forecast steel price using time series analysis method. A review of the research carried out in the area of steel forecasting and Machine Learning helped in the generation and refinement of a focused problem statement. From the research gap, it was found that time series model-SARIMAX and FB Prophet will be the best model for forecast of steel price and also it was decided to collect a larger dataset than previous study. A questionnaire was prepared and semi-structured interview was conducted with various stakeholders like developers, contractors, structural consultants, steel vendors, etc. for collection of data and getting knowledge about the indicators which could be considered for the study. Based on the insights from them, the data for daily steel price and the data for lead indicators like iron ore, aluminum, coal, crude oil, inflation, GDP, etc. was collected from 2003 to 2022. Once the data collection was completed, several steps such as data cleansing, transforming, extrapolating, and integrating helped to prepare a coherent dataset fit for training the models. The exploratory data analysis helped in discovering patterns, outliers, abnormalities and eliminating null values. Three forecasting models were prepared- SARIMAX (univariate model) and Prophet (univariate model and multivariate model). For training the model, the data from Jan 2003 till Jun 2022 was considered and for testing the model, the data from July 2022 to Dec 2022 was taken into account. The forecast of steel price was done for 6 months – Jul 2022 to Dec 2022. From the Mean Absolute Percent Error (MAPE), the most accurate model was recognized and the forecasted steel price was compared with the actual steel price.