Comparison of different machine learning techniques in river flow prediction

Water has a complex relationship with agricultural activities, economy, health, use of energy resources andhygiene. In parallel with climate change and population growth, the inadequacy of our water resources in thecoming years is one of the major problems we are likely to face. Estimating how much of the available water canbe used and how much the water potential will change in the future can enable more accurate water planning. Theaverage flow, total precipitation and average temperature values of Tagar Stream K & uuml;& ccedil;& uuml;kkumluk and Porsuk StreamPorsuk & Ccedil;iftli & gbreve;i Station were analyzed together and the river flow rate was estimated. Linear Regression, SupportVector, Decision Trees, Random Forest and Extra Trees methods were applied for flow estimation. In order to improve the performance of each of the applied models, a hybrid method was developed using Wavelet Transform.Approximately 65% of the dataset is divided into training, 15% validation and 20% test data. ETR was found to bethe most successful prediction method with 70.8% for Tagar Stream K & uuml;& ccedil;& uuml;kkumluk and 67.67% for Porsuk StreamPorsuk Farm. The developed hybrid model increased the success rate for all methods; the highest increase wasobtained in SVR method with 20.82% for K & uuml;& ccedil;& uuml;kkumluk and 30.64% for Porsuk Farm. The most successful methodwas obtained in the ETR method with 91.46% and 86.39%.