This paper employs machine learning algorithms with mathematically designed candlestick pattern in attempting to create a five-step procedure that examines every movement in the stock market in an effort to put every move of a stock into the precise buy, sell, and hold signals. Comprehensive exploratory data analysis ensures data purity and identification and categorization of all 68 candlestick patterns as neutral, bearish, and bullish is definitely the start. Careful class imbalance management through SMOTE and Tomek Links in feature engineering and preprocessing in the second step addresses encoding and scaling carefully. For generalization, the third step is training a stacking classifier model with tenfold cross-validation using the ensemble of SVC, GNB, LR and DT as base models and Random Forest Regressor as a meta-model. Model performance was always quantified in terms of overall recall, overall precision, overall F1-score and overall confusion matrix on all correct predictions. Prediction model is also applied in the last step and data generates algorithmic trading and portfolio management trade signals. Machine learning and feature engineering are used in the system to effectively forecast most of the stock trends, highly improving the portfolio performance and lowering the risk immensely through trading risky financial markets