Using Artificial Neural Networks to Optimize Acceleration Due to Gravity g Measurement in a Compound Pendulum Experiment

Abstract: In this study, we explore a novel approach of implementing the Artificial Neural Network (ANN) model to validate a traditional experiment performed in the undergraduate physics laboratory, e.g. measurement of acceleration due to gravity g using compound pendulum. The input layer for the ANN model comprises of different parameters of the compound pendulum experiment such as its effective length, time period of oscillation and initial angular displacement. The model is first trained by using 70 percent of the experimental data. Then the trained model is validated and tested on the rest 30 percent of the experimental data which are treated as unseen data to predict the value of g. The ANN-predicted values are compared with the experimental value of g to assess precision. The average value of g was determined to be 1009.029797cm/s^2 with a random error of $\pm$6.817633 cm/s^2 through traditional experiment. However, the ANN-predicted value of g was 1009.029858 with a mean absolute error of 0.000592 cm/s^2. The authors propose that ANN-based methodologies could bridge the gap between theoretical understanding and practical application by introducing students to cutting-edge computational techniques. Such integration allows for a deeper engagement with the subject matter, encouraging critical thinking and problem-solving skills in experimental physics. This innovative approach underscores the transformative role of machine learning in modern physics education. The model's performance demonstrates high accuracy and robustness in predicting g, outperforming traditional empirical approaches. Our results indicate that ANN-based optimization can significantly improve the precision of gravitational measurements, offering a reliable and efficient tool for applications that require high-accuracy determination of gravitational acceleration.