Unsupervised Climate Anomaly Detection with IF and OCSVM

Long-term changes in temperature allow us to identify regional climate variability and notice early warnings of environmental changes. Unsupervised machine learning methods were applied to monthly temperature differences in two critical climate-sensitive locations in northwest Bangladesh. The analysis here uses monthly data collected from 1980 to 2024. The Z-score method was first applied to the data values to identify anomalies. Without access to labeled data, IF and OCSVM were used as unsupervised machine learning algorithms to spot outliers in the data. The method identified acute, one-time temperature outliers in the isolation forest, while OCSVM detected gradual changes occurring over a period. Each model distinguished a specific set of years, which comprised 1998, 2013, and 2024. While both sites experience thermal and other anomalies, the data show that Ishwardi is gradually developing them, and Rajshahi has more intense thermal spikes. Rising regional climate change and continued warming are evident in the numerous sharp temperature spikes that occurred after 2000. It discusses the use of unsupervised machine learning to identify unusual climate changes. Besides making detections more reliable, using both modes from the satellite helps give a clearer view of the local weather conditions. The results of this study will support environmental risk analysis and supply facts to guide upcoming climate tracking.