A 15-year (1999-2014) homogeneous database from RSMC La Reunion is employed to examine the large-scale characteristics of rapidly intensifying tropical cyclones over the southwest Indian Ocean. To allow for basin inter-comparisons we followed the methodology used by Kaplan and DeMaria (2003; 2010) for Atlantic systems. Rapid intensification (RI) for oceanic tropical systems is statistically defined as a 24-hour intensity change of the maximum surface winds exceeding 15.4 m s-1 (94.4th percentile of the ΔV24 distribution). This is equivalent to the 30-kt official threshold determined for the North Atlantic basin (94th percentile) using 1 min sustained winds, whereas the present study uses WMO standard 10 min average winds. It is shown that 39% of all the 151 tropical systems, 12% that reached the tropical storm stage (34 kts), 73% that reached the tropical cyclone stage (64kts), and all the very intense tropical cyclones (115 kts) underwent RI at least once during their lifetimes. Interestingly, the RI probability distribution as a function of storm initial wind is bimodal with two peaks in the 40-55 kts and 60-75 kts ranges. To better understand and ultimately predict RI, the ERA-Interim dataset is used to determine the large-scale dynamic and thermodynamic conditions that are conducive for RI. Six environmental parameters averaged along the storm track during the 24-h period following the beginning of RI are identified as potential RI predictors. For each of them, statistically significant differences are found between the mean values of the RI and non-RI samples at the 99.9% level using a two-sided t test. RI predictors are, in decreasing order of importance: TC intensification during the previous 12h (DVMXM12), a high upper-level divergence (DIV200), a weak 850-200-hPa vertical wind shear (SHR), a high sea surface temperature in a 200-km radius surrounding the storm center (SST), a weak upper-level cyclonic potential vorticity (on the 350-K isentrope, PV350). A statistical-dynamical tool aiming at estimating the probability of RI over the next 24 h is also developed for the southwest Indian Ocean using best multilinear regression of the most relevant environmental RI predictors.