<p>This study uses the k-nearest neighbor (KNN) algorithm, Pearson correlation coefficient to predict sudden storm commencement (SSC), main, and recovery phases of ten intense storms (occurred during 2012-2018), and solar wind coupling with Earth’s magnetosphere. Probability distribution functions (pdfs) for solar wind interplanetary magnetic field (IMF), its Bz component and solar wind speed are also studied. IMF-Bz pdfs included single peak with and without bumps in front and on tail, double peak, and triple peak but highest positive correlation coefficients for double peak dominated those for triple peak distributions. Similar results are found for IMF pdfs. Contrarily, highest positive correlation coefficient occurred for single peak solar wind probability distribution. It is also found that storm phase prediction accuracy reduced when number of nearest neighbors is increased. The accuracy of prediction changed by replacing elements of data vectors. The highest accuracy occurred for data vectors including SYMH index and Bz component of an interplanetary magnetic field. The findings of this study can play a critical role in future space weather.</p>