Abstarct: In this study, I present an automated frxamework for classifying software reports using machine learning algorithms and uncertainty calibration. The study addresses the challenge of manually classifying software reports by automating the classification process. The dataset used in this research was the Bugzilla software bug reports dataset, with the main experiments focusing on the Fix Classification task. The primary objective was to classify reports according to whether they represented valid and fixable software defects requiring an actual code modification, combined with uncertainty calibration. The study addresses the challenge of manually triaging software bug reports by allowing the automation of the classification of these bug reports. The dataset used for this research was the Bugzilla Bug Reports dataset. The primary focus of the research was the task of predicting the resolution status of bug reports—specifically, classifying whether a report represents a valid defect requiring an actual code fix, baxsed on its textual content. The text fields within the bug report (tixtle and descxription) were pre-processed or cleaned, combined and truncated; then these text fields were transformed into numerical features for use in building various classical machine learning models using the Term Frequency-Inverse Document Frequency (TF-IDF) method and then tokenized for building the BiLSTM (Bidirectional Long Short-Term Memory) model. A number of classifiers were developed and compared (classifiers include Logistic Regression, Linear Support Vector Machine, Multinomial Naive Bayes, Complement Naive Bayes and BiLSTM). In order to enhance classifier performance and to consistently produce reliable confidence estimates on the model predictions, techniques such as hyperparameter tuning, sample weighting, threshold tuning and calibration techniques were applied. Additionally, calibration methods for all the classical machine learning models included Sigmoid calibration, and for the BiLSTM model calibration methods included temperature scaling and Monte Carlo Dropout. The classifiers were evaluated using various metrics, including accuracy, precision, recall, weighted F1-score, confusion matrix, expected calibration errors, Brier Scores, reliability diagrams, and a low-confidence case analysis. Ultimately, the results of this study showed that the TF-IDF Logistic Regression with Sigmoid calibration produced the best overall classification performance of the classifiers used by achieving an overall accuracy and weighted F1-score of 85.52% and 85.60%, respectively.