Five years of data from 2001 until 2006 of warm unstratified shallow, oligotrophic to mesothropic tropical Putrajaya Lake, Malaysia were used to study pattern discovery and forecasting of the diatom abundance using supervised and unsupervised artificial neural networks. Recurrent artificial neural network (RANN) was used for the supervised artificial neural network and Kohonen Self Organizing Feature Maps (SOM) was used for unsupervised artificial neural network. RANN was applied for forecasting of diatom abundance. The RANN performance was measured in terms of root mean square error (RMSE) and the value reported was 29.12 cell/mL. Classification and clustering by SOM and sensitivity analysis from the RANN were used to reveal the relationship among water temperature, pH, nitrate nitrogen (NO3-N) concentration, chemical oxygen demand (COD) concentration and diatom abundance. The results indicated that the combination of supervised and unsupervised artificial neural network is important not only for forecasting algae abundance but also in reasoning and understanding ecological relationships. This in return will assist in better management of lake water quality.