With the development of deep learning and availability of large datasets, the evaluation of player performance has expanded the scope of sports analytics. Predicting the outcomes of players is still of utmost importance for tactical decisions. While Ridge Regression, Gradient Boosting, and Support Vector Regression (SVR) and other traditional machine learning techniques used in sports analytics, the true potential of the complex temporal dependencies of player statistics is still not achieved. These temporal dependencies in the performance data can inspire the design of advanced techniques, so this paper proposes a hybrid RNN-LSTM architecture suitable for player performance evaluation. The proposed model utilizes the forecasting abilities of RNNs coupled with LSTM cells to synthesize memory and guarantee performance. The simulation involved a dataset containing 1007 player-season records spanning 2008 to 2024 and including various bowling and batting performance metrics. All model training and evaluation were performed in the TensorFlow and Keras libraries. The experimental results show that our proposed RNN-LSTM has an accuracy of 0.9635, R2 score of 0.9910, Mean Square Error (MSE) of 301.4782, and Mean Absolute Error (MAE) of 7.6366 that outperform Ridge Regression, Gradient Boosting and SVR baselines. In the end, a hybrid RNN-LSTM model will be highly accurate and scalable. More importantly, it will set the benchmark for player performance prediction. It can predict better than regular machine learning techniques because it captures temporal dependencies.