The invention relates to an online anomaly detection method for a big data platform based on a maximum information coefficient. The method adopts a Spark Streaming sliding window mechanism to perform statistical analysis on an access log, constructs a user activity model, so as to calculate the maximum information coefficient of a user activity sequence of the adjacent time windows, and determines whether the current time has a network anomaly by comparing the maximum information coefficient with a set threshold. The method is beneficial to find the network anomalies under online scenes quickly, accurately and in near real time, and realizes the near real-time anomaly detection function of an anomaly detection and analysis system of the big data platform.