This paper discusses the multilevel approach in constructing a model for estimating hierarchically structured data of students' performance. Multilevel models that take into account variation from the clustering of data in different levels are compared to regression models using least squares method. This study also estimates the contributions of gender and ethnic factors on students' performance. Performance data of866 students in a science faculty in an institution of higher learning is obtained and analyzed. This data is hierarchically structured with two levels, namely students and departments. Analysis findings show different parameter estimates for both models. Also, the multilevel model which incorporates variability from different levels and predictors from higher levels is found to provide a better fit for model explaining students' performance.
[Rencana ini membincangkan pendekatan multitahap dalam pembinaan model penganggaran pencapaian pelajar yang mempunyai struktur data hierarki. Model multitahap yang mengambil kira variasi data yang berpunca dari pengelompokan data pada tahap-tahap yang berbeza dibandingkan dengan model regresi linear yang menggunakan kaedah kuasa dua terkecil. Seterusnya kajian ini menganggar sumbangan faktor jantina dan etnik ke atas pencapaian pelajar. Data pencapaian akademik seramai 866 pelajar fakulti sains di sebuah institusi pengajian tinggi telah diperoleh dan dianalisis. Data pelajar ini berstruktur hierarki dengan dua tahap, iaitu pelajar dan jabatan. Hasil kajian menunjukkan kedua-dua kaedah memberikan penganggaran yang berbeza. Malah, didapati model multitahap yang memasukkan variasi dari tahap-tahap berlainan dan pembolehubah peramal dari tahap yang lebih tinggi memberikan padanan model lebih baik bagi menerangkan pencapaian pelajar].
Given the carcinogenic properties of formaldehyde-based chemicals, an alternative method for resin-finishing cotton textiles is urgently needed. Therefore, the primary objective of this study is to introduce a sustainable resin-finishing process for cotton fabric via an industrial procedure. For this purpose, Bluesign® approved a formaldehyde-free Knittex RCT® resin was used, and the process parameters were designed and optimized according to the Taguchi L27 method. XRD analysis confirmed the crosslinking formation between resin and neighboring molecules of cotton fabric, as no change in the cellulose crystallization phase. Several machine learning models were built in a sequence to predict the crease recovery angle (CRA), tearing strength (TE) and whiteness index (WI). Assessment of modelling was evaluated through the use of various metrics such as root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R2). Results were compared to those from other regression models, such as principal component regression (PCR), partial least squares regression (PLSR), and fuzzy modelling. Based on the results of our research, the LSSVR model predicted the CRA, TE, and WI with substantially more accuracy than other models, as shown by the fact that its RMSE and MAE values were significantly lower. In addition, it offered the greatest possible R2 values, reaching up to 0.9627.