Nowadays, there is a big increase in the usage ofdata analytic applications and services because of the growth inthe data produced from many different sources. The QoSproperties such as response time, reliability and latency of theseservices are important factors to decide which services to select.As we know, the energy consumption is becoming a big issue as aresult of IT expansion. Therefore, establishing a QoS-based webservice selection approach that considers energy consumption asone of the essential QoS properties represents a significant steptowards selecting the greener web service. This paper presents anexperimental study of energy consumption and latency behaviorof data mining algorithms running as web services. Our studyshows that, there is a strong relation between the datasetproperties such as dataset size, number of attributes, data type,and QoS attributes energy consumption and latency. Based onthe findings from our study, a prediction system is built whichcan be used to predict the energy consumption and latency valuesfor data mining web services on a given dataset, and then theseservices can be ranked according to their predicted energy andlatency values. Experimental results show the effectiveness of ourprediction and service selection system.