Sentiment analysis has emerged to acknowledge public opinion, customer preferences, and societal trends through textual data. Sentiment analysis problem has previously been solved by various machine learning and deep learning tools but less attention was given on the outliers and dimensionality reduction. Due to large amount of data generated, it is essential to reduce the dimension of the data and outliers present in the dataset. This paper presents a framework for improved sentiments analysis using Principle Component Analysis (PCA) for reducing the dimension, K-means clustering combined with BERT for robust classification. K-means clustering is employed to group parallel data points facilitating outlier removal and better organization of the dataset. The framework is validated on the maximum-scale Twitter dataset, showing significant improvements in performance and processing speed compared to conventional methods.