Abstract: Social media has reshaped the way in which individuals communicate with disruptive negative and positive externalities, alongside being breeding grounds for bot activities. Social media, limited to Twitter in this research, is rarely monitored and has accelerated the dissemination of disinformation, most of which is attributed to bot activities. To date, bot detection techniques are challenged by their inability to keep abreast with bot progression, while its importance is increasingly being bought to the forefront of political discussions. Therefore, the research will introduce a local-global SVM (LG-SVM) model to increase flexibility in detection methods to keep abreast with the ever-evolving world of Twitter. The end- to-end framework has increased accuracy in detection by treating the data as unsupervised and introducing an autocoder to reduce dimensionality, which has solved for over-fitting. The multi-level framework starts with a convolutional layer that reduces dimensionality and learns key short-term characteristics, followed by a max pooling layer, two bi-LSTM models, a bottleneck layer, an upscaling layer, and a deconvolution layer. Dimension reduction oversaw the reduction of the dataset into 3 lengths: 70, 100, and 130, resulting in a 99% accuracy rate. Additionally, a multilayer perceptron model is built to predict the impact Bot Tweets have on stock market volatility with a 96.53% coefficient correlation.