Semiconductor supply chain is interlocking. Any delays in the process will affect the whole supply chain including product delivery. Through data stream mining, the study focuses on classification on the degree of on-time delivery, and to analyze the effect of different drifts and drift detection techniques. Data streams have been generated based on a simulated scenario of semiconductor supply chain in Anylogic. The study utilized WEKA and Massive Online Analysis (MOA) software to perform data stream processing and classification. Data preprocessing was performed via WEKA to ease the classification. Sudden drift and gradual drift were simulated using MOA. For classification, Naïve Bayes used as base learner and different drift detection techniques such as CUSUM test, Page-Hinkley test, DDM, EDDM, ADWIN have been used to analyze its effectiveness towards the simulated drifts. Without using drift detection technique, the classification accuracy dropped after introducing drift to the stream. By using drift detection technique, most classification accuracies have been improved. Each drift detection technique performed differently on the types of drift. In terms of overall classification accuracy, all the selected drift detection techniques performed better in gradual drift than sudden drift. By amending the sensitivity parameters of CUSUM test and Page-Hinkley Test, accuracy would be affected. While for ADWIN, it might require a slightly larger delta value to amend its sensitivity towards the sudden drift.