Analysis of the Effect of Data Scaling on the Performance of the Machine Learning Algorithm for Plant Identification

Abstract

Data scaling has an important role in preprocessing data that has an impact on the performance of machine learning algorithms. This study aims to analyze the effect of min-max normalization techniques and standardization (zero-mean normalization) on the performance of machine learning algorithms. The stages carried out in this study included data normalization on the data of leaf venation features. The results of the normalized dataset, then tested to four machine learning algorithms include KNN, Naïve Bayesian, ANN, SVM with RBF kernels and linear kernels. The analysis was carried out on the results of model evaluations using 10-fold cross-validation, and validation using test data. The results obtained show that Naïve Bayesian has the most stable performance against the use of min-max normalization techniques as well as standardization. The KNN algorithm is quite stable compared to SVM and ANN. However, the combination of the min-max normalization technique with SVM that uses the RBF kernel can provide the best performance results. On the other hand, SVM with a linear kernel, the best performance is obtained when applying standardization techniques (zero-mean normalization). While the ANN algorithm, it is necessary to do a number of trials to find out the best data normalization techniques that match the algorithm.

Effect of Data Scaling Methods on Machine Learning Algorithms and Model Performance

“…While there are several data scaling techniques available, one of the main challenges associated with ML is to choose the appropriate scaling method. Many studies bolster the effect of data scaling techniques on different ML algorithms [31,32]. Shahriyari et al (2019) showed that the performance of normalization has a significant effect on different ML approaches [32].…”

“…Another study conducted by Ambarwari et al (2020) showed that data scaling techniques such as MinMax normalization and standardization have also significant effects on data analysis [31]. The study was carried out using ML algorithms such as KNN, Naïve Bayesian, ANN, and SVM with RBF.…”

An Investigation on Disparity Responds of Machine Learning Algorithms to Data Normalization Method

“…The majority of the works in the literature that have investigated the effect of the min-max data normalization reported a positive impact of the min-max normalization on the adopted ML techniques in their studies (Dadzie and Kwakye, 2021;Shahriyari, 2017), while some other studies determined that its usefulness varies from good to bad depending on the nature of the datasets and the ML model (Ambarwari, Adrian, and Herdiyeni, 2020) (Ahsan, et al, 2021). On the other hand, very limited studies concluded the degradation of the ML model accuracy with the present of min-max normalization.…”

“…Many research works showed that SVM has more sensitive responds than ANN and E-KNN toward the normalization techniques. Results in many SVM-based works showed the usefulness of using min-max normalization with SVM (Dadzie and Kwakye, 2021;Shahriyari, 2017;Ambarwari, Adrian, and Herdiyeni, 2020). Despite that, there are still some studies proved that normalization has no effect or has very little effect on the accuracy rate of the SVM-based models (Singh, et al, 2015).…”

“…Some works showed the negative effect of min-max normalization (Singh, et al, 2015), whereas others concluded that no significant performance improvements were observed (Jayalakshmi and Santhakumaran, 2011). No improvements were also observed for higher-order neural networks (Prasetyo, Setiawan, and Adji, 2020) and deep learning algorithms (Pires et al, 2020), while other research works reported an obvious improvement in the accuracy of the models (Ambarwari, Adrian, and Herdiyeni 2020).…”