<p>Predicting battery capacity is essential for enhancing battery management systems (BMSs), ensuring safety, and extending battery life. However, lithium-ion battery faces the challenge of performance degradation over the period due to electrochemical phenomena. It can be addressed with data-driven techniques to estimate the battery capacity and remaining useful life (RUL). The machine learning (ML) algorithm efficacy directly impacted by the data types. NASA and CALCE datasets are used to validate the applicability of ML algorithms. The dataset are divided into training and testing sets based on charged-discharged cycles. Pretraining datasets are tested in time series with forward prediction judgment of the data size to predict RUL. There may be an overfitting or underfitting problems in estimating capacity of the battery. However, such problems can be addressed with proper tuning of hyperparameters in time series model with, number of trees, maximum depth of the tree and splitting the data points. As data may be noisy or nonlinear, in most cases, RF prevents overfitting or underfitting by building multiple decision trees which reduces the variance and increases accuracy. RF achieves comparable prediction accuracy, even when trained on limited data as compared to existing data-driven techniques in terms of error metrics RMSE, MSE, MAPE, and <i>R</i>². The findings highlight RF as a preferred choice with an average RMSE error reduced to 5.66E-16 and predict the battery RUL maximum error in four cycles to lithium-ion battery. These techniques may provide robustness to BMS in real-time applications.</p>