In this paper, authors have proposed a technique which uses the existing database of chess games and machine learning algorithms to predict the game results. Authors have also developed various relationships among different combinations of attributes like half-moves, move sequence, chess engine evaluated score, opening sequence and the game result. The database of 10,000 actual chess games, imported and processed using Shane’s Chess Information Database (SCID), is annotated with evaluation score for each half-move using Stockfish chess engine running constantly on depth 17. This provided us with a total of 8,40,289 board evaluations. The idea is to make the Multi-Variate Linear Regression algorithm learn from these evaluation scores for same sequence of opening moves and game outcome, then using it to calculate the winning score of a side for each possible move and thus suggesting the move with highest score. The output is also tested with including move details. Game attributes are also classified into classes. Using Naïve Bayes classification, the data result is classified into three classes namely move preferable to white, black or a tie and then the data is validated on 20% of the dataset to determine accuracies for different combinations of considered attributes.