Comparative Analysis of Data Normalisation Techniques: A Case of Reservoir Water Level Monitoring


Ishak W. H. W., Yamin F. M., Maidin S. S., Ajibade S. M.

Journal of Soft Computing and Data Mining, cilt.7, sa.3, ss.28-38, 2026 (Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 7 Sayı: 3
  • Basım Tarihi: 2026
  • Doi Numarası: 10.30880/jscdm.2026.07.03.003
  • Dergi Adı: Journal of Soft Computing and Data Mining
  • Derginin Tarandığı İndeksler: Scopus
  • Sayfa Sayıları: ss.28-38
  • Anahtar Kelimeler: Data normalisation, Decimal Scaling, Min-Max, neural network, reservoir monitoring, Z-score
  • İstanbul Ticaret Üniversitesi Adresli: Evet

Özet

Reservoir water level data may vary with seasonal rainfall, reservoir operations and extreme hydrological events. This study compares four normalisation techniques applied to 3,287 daily observations from the Timah Tasoh Reservoir between 1 January 1999 and 31 December 2007: Min-Max (binary), Min-Max (bipolar), Z-score and Decimal Scaling. The original water levels ranged from 26.81 to 30.00 m, with a skewness of −0.878 and an excess kurtosis of 0.327. Using descriptive statistics and Pearson correlation, we compared the resulting scales and statistical characteristics of the transformed observations. Strengths, weaknesses, opportunities and threats (SWOT) analysis was also used to summarise the practical characteristics of each technique. The study identified 23 observations meeting the |Z| > 3 outlier criterion. Min-Max normalisation mapped the observed range to 0–1 or −1 to 1, while Decimal Scaling divided each water level by 100. All four transformations had a reported Pearson correlation of 1.000 with the original observations, consistent with their positive linear formulas. The techniques produced different scales but preserved the ordering of observations. No forecasting algorithm was trained or tested, so the results do not establish which technique would yield the best predictive accuracy. The comparison provides statistical information to guide preprocessing choices for future reservoir forecasting studies.