Perbandingan Kinerja Algoritma Ridge dan LASSO Regression dalam Prediksi Persentase Lemak Tubuh

Abstract

Body Fat Percentage (BFP) is an accurate measurement for diagnosing obesity, the main factor causing obesity is an energy imbalance between calorie expenditure and intake. People who are obese are more likely to experience diabetes, heart disease, stroke, and musculoskeletal problems. According to the World Health Organization (WHO), more than 1.9 billion people aged 18 and over were overweight in 2016, and more than 650 million of them were obese. The BFP value is very important for determining the likelihood of someone being obese, but measuring it can be challenging, expensive, and painful. To predict a person's BFP value based on their body size, this study used data mining techniques, specifically the Ridge Regression and LASSO Regression algorithms. Both of these algorithms use regularization techniques, namely by adding a penalty to the coefficient value towards zero. The data used amounted to 250 rows, which were divided into two parts based on the percentage of 80:20 of the total data. A total of 200 rows of data (80%) were used as training data, while 50 rows of data (20%) were used as test data. The results of this study indicate that the LASSO Regression algorithm is superior for the entire evaluation matrix, with an MSE value of 22.56, MAE of 3.64, R-squared of 0.60, and Adjusted R-squared of 0.44, inversely proportional to Ridge Regression with an MSE value of only 22.89, MAE of 3.80, R-squared of 0.59, and Adjusted R-squared of 0.43.