This study presents the design and development of a low-cost near infrared (NIR) spectroscopy system using NIR LEDs and regression models. The proposed system effectively measures shortwave NIR spectrum while eliminating unwanted signals, achieving a root mean squared error of 1.1616nm for wavelength estimation. Cross-validation results indicate that the quadratic regression model provides the best performance, validating the feasibility of the proposed design for various NIR spectroscopic applications.