Security-Constrained Unit Commitment (SCUC) is a fundamental problem in power\nsystems and electricity markets. In practical settings, SCUC is repeatedly\nsolved via Mixed-Integer Linear Programming, sometimes multiple times per day,\nwith only minor changes in input data. In this work, we propose a number of\nmachine learning (ML) techniques to effectively extract information from\npreviously solved instances in order to significantly improve the computational\nperformance of MIP solvers when solving similar instances in the future. Based\non statistical data, we predict redundant constraints in the formulation, good\ninitial feasible solutions and affine subspaces where the optimal solution is\nlikely to lie, leading to significant reduction in problem size. Computational\nresults on a diverse set of realistic and large-scale instances show that,\nusing the proposed techniques, SCUC can be solved on average 4.3x faster with\noptimality guarantees, and 10.2x faster without optimality guarantees, but with\nno observed reduction in solution quality. Out-of-distribution experiments\nprovides evidence that the method is somewhat robust against dataset shift.\n