System and method for predicting defects in a computer program

A method for predicting defects in a computer program during its development, the method comprising:

• applying, by a processor, a machine learning algorithm including a Frequent Pattern Growth (FP-Growth) algorithm to one or more sets of source code files;
• identifying the among the one or more sets of source code files one or more frequent sets of source code files that have been modified together in a plurality of historical builds associated with the computer program;
• identifying one or more frequent set of defect keywords that have been logged together in the plurality of historical builds associated with the computer program, wherein each set of the one or more frequent set of defect keywords comprises a group of defect keywords that have been recorded together in the one or more historical builds, further wherein the defect keywords that have been logged together with a predefined minimum frequency value are considered to be a part of the frequent set of the defect keywords;
• establishing one or more patterns between the one or more frequent sets of modified source code files and the one or more frequent set of defect keywords by identifying co-occurrence between the one or more frequent sets of modified source code files and the one or more frequent sets of defect keywords for each historical build;
• receiving data associated with one or more new builds associated with the computer program;
• and
• predicting one or more defects in the one or more new builds using the one or more established patterns, wherein the predicted or more defects comprises:
• matching the one or more source code files that have been modified in the one or more new builds with the one or more frequent sets of modified source code files identified from the established one or more patterns, wherein a match signifies a likelihood of receiving one or more defects in the new build,
• wherein, for every occurring match, the processor filters the defects based on the corresponding identified sets of frequent defect keywords, and recommends the filtered defects for testing;
• wherein the processor is trained to learn historical data patterns known to cause a previous defect in a given historical build among the plurality of historical builds of the software program, the learned historical data patterns including at least one of patterns between one or more source files and one or more failure test cases, between user story keywords and one or more defect keywords, and between user story keywords and one or more failure test cases, and following the training is configured to predict the one or more defects associated with the one or more new builds based on the previous defect learned from the historical data patterns.

A computer program product comprising:

• a non-transitory computer-readable medium having computer-readable program code stored thereon, the computer-readable program code comprising instructions that when executed by a processor, cause the processor to:
• identify one or more frequent sets of source code files that have been modified together in plurality of historical builds associated with the computer program;
• identify one or more frequent set of defect keywords that have been logged together in the plurality of historical builds associated with the computer program, wherein the defect keywords that have been logged together with a predefined minimum frequency value are considered to be a part of the frequent set of the defect keywords;
• establish one or more patterns between the one or more frequent sets of modified source code files and the one or more frequent set of defect keywords by identifying co-occurrence between the one or more frequent sets of modified source code files and the one or more frequent sets of defect keywords for each historical build;
• receive data associated with one or more new builds associated with the computer program;
• and
• predict one or more defects in the one or more new builds using the one or more established patterns,
• wherein the processor is trained to learn historical data patterns known to cause a previous defect in a given historical build among the plurality of historical builds of the software program, the learned historical data patterns including at least one of patterns between one or more source files and one or more failure test cases, between user story keywords and one or more defect keywords, and between user story keywords and one or more failure test cases, and following the training is configured to predict the one or more defects associated with the one or more new builds based on the previous defect learned from the historical data patterns.