Performance Of Developed Model Research Articles

Ensemble Learning Approach for Developing Performance Models of Flexible Pavement

This research utilizes the Long-Term Pavement Performance database, focusing on devel-oping a predictive model for flexible pavement performance in the Southern United States. Analyzing 367 pavement sections, this study investigates crucial factors influencing asphaltic concrete (AC) pavement deterioration, such as structural and material components, air voids, compaction density, temperature at laydown, traffic load, precipitation, and freeze–thaw cycles. The objective of this study is to develop a predictive machine learning model for AC pavement wheel path cracking (WpCrAr) and the age at which cracking initiates (WpCrAr) as performance indicators. This study thoroughly investigated three ensemble machine learning models, including random forest, extremely randomized trees (ETR), and extreme gradient boosting (XGBoost). It was observed that XGBoost, optimized using Bayesian methods, emerged as the most effective among the evaluated models, demonstrating good predictive accuracy, with an R2 of 0.79 for WpCrAr and 0.92 for AgeCrack and mean absolute errors of 1.07 and 0.74, respectively. The most important features influencing crack initiation and progression were identified, including equivalent single axle load (ESAL), pavement age, number of layers, precipitation, and freeze–thaw cycles. This paper also showed the impact of pavement material combinations for base and subgrade layers on the delay of crack initiation.

Enhancing rutting depth prediction in asphalt pavements: A synergistic approach of extreme gradient boosting and snake optimization

This study investigates the development of pavement performance models, focusing on the predictive analysis of pavement distress using ensemble machine learning methods. We employ the Research Institute of Highway Ministry of Transport track dataset to construct predictive models for pavement distress, comparing various methodologies including Random Forest (RF) regression, Extreme Gradient Boosting (XGB) machines, Snake Optimizer-Random Forest (SO-RF), and Snake Optimizer-Extreme Gradient Boosting (SO-XGB). The approach is conducted by multiple experiments and a comprehensive comparative analysis leveraging the repeated K-Fold cross-validation method to ensure the robust assessment for these models. Our findings show that the SO-XGB method is particularly effective in predicting rutting depth. The strength of the SO-XGB model lies in its excellent performance in several performance metrics, such as coefficient of determination (R2), the mean absolute error (MAE), root mean square error (RMSE), the mean absolute percentage error (MAPE), performance index (PI), index of agreement (IA), variance accounted for (VAF) and objective function (OBJ). These results not only illustrate the model’s accuracy but also the practical applicability of SO-XGB model in automatically forecasting pavement damage. Thus, the SO-XGB method stands out as a direct, powerful, and efficient tool in the field of pavement performance prediction.

Use of Human-Automation Taxonomies for System Modeling

Skraaning and Jamieson indicate the importance of developing a definition of automation failure in order to guide development of human performance models to predict human-automation interactions. They go on to develop a taxonomy including four categories: Elementary automation failures; Systemic automation failures; Human-automation breakdowns; and Human and organizational slips and misconceptions. They suggest that the latter category be excluded from classes of automation failures. They then proceed to identify important factors within each category. This paper presents a complementary taxonomy designed to identify factors contributing to automation-related failures in order to develop probes to elicit predictions of failure stories for some proposed new system design from a range of experts. While there is overlap between these two taxonomies, there are important differences. It is suggested that they could be integrated or used jointly both for knowledge elicitation and to identify requirements regarding needed capabilities of predictive performance models.

Development of performance and learning rate evaluation models in robot-assisted surgery using electroencephalography and eye-tracking

The existing performance evaluation methods in robot-assisted surgery (RAS) are mainly subjective, costly, and affected by shortcomings such as the inconsistency of results and dependency on the raters’ opinions. The aim of this study was to develop models for an objective evaluation of performance and rate of learning RAS skills while practicing surgical simulator tasks. The electroencephalogram (EEG) and eye-tracking data were recorded from 26 subjects while performing Tubes, Suture Sponge, and Dots and Needles tasks. Performance scores were generated by the simulator program. The functional brain networks were extracted using EEG data and coherence analysis. Then these networks, along with community detection analysis, facilitated the extraction of average search information and average temporal flexibility features at 21 Brodmann areas (BA) and four band frequencies. Twelve eye-tracking features were extracted and used to develop linear random intercept models for performance evaluation and multivariate linear regression models for the evaluation of the learning rate. Results showed that subject-wise standardization of features improved the R2 of the models. Average pupil diameter and rate of saccade were associated with performance in the Tubes task (multivariate analysis; p-value = 0.01 and p-value = 0.04, respectively). Entropy of pupil diameter was associated with performance in Dots and Needles task (multivariate analysis; p-value = 0.01). Average temporal flexibility and search information in several BAs and band frequencies were associated with performance and rate of learning. The models may be used to objectify performance and learning rate evaluation in RAS once validated with a broader sample size and tasks.

Using machine learning to classify temporomandibular disorders: a proof of concept.

the escalating influx of patients with temporomandibular disorders and the challenges associated with accurate diagnosis by non-specialized dental practitioners underscore the integration of artificial intelligence into the diagnostic process of temporomandibular disorders (TMD) as a potential solution to mitigate diagnostic disparities associated with this condition. In this study, we evaluated a machine-learning model for classifying TMDs based on the International Classification of Orofacial Pain, using structured data. Model construction was performed by the exploration of a dataset comprising patient records from the repository of the Multidisciplinary Orofacial Pain Center (CEMDOR) affiliated with the Federal University of Santa Catarina. Diagnoses of TMD were categorized following the principles established by the International Classification of Orofacial Pain (ICOP-1). Two independent experiments were conducted using the decision tree technique to classify muscular or articular conditions. Both experiments uniformly adopted identical metrics to assess the developed model's performance and efficacy. The classification model for joint pain showed a relevant potential for general practitioners, presenting 84% accuracy and f1-score of 0.85. Thus, myofascial pain was classified with 78% accuracy and an f1-score of 0.76. Both models used from 2 to 5 clinical variables to classify orofacial pain. The use of decision tree-based machine learning holds significant support potential for TMD classification.

Machine learning based pavement performance prediction for data-driven decision of asphalt pavement overlay

This study is to develop pavement performance models using support vector regression and ensemble machine learning methods for selection of pavement overlay strategy. Predictive models of pavement distresses were developed based on the Long-Term Pavement Performance (LTPP) data and compared using support vector machine, random forest regression, gradient boosting machine, and stacking ensemble. Gradient boosting machine was found to be a more effective method to establish predictive models of rut depth and International Roughness Index. Stacking ensemble and random forest regression would provide reliable prediction of alligator cracking. The models developed based on the clusters of climate and traffic parameters were found to be more effective. Based on the developed performance models, the effects of asphalt overlays on pavement distresses and service lives were investigated. When the overlay with recycled asphalt concrete (AC) was applied, the propagation of alligator cracking was faster compared to the overlay with virgin asphalt mixture. Milling before overlay tended to slow the increase of IRI but fasten the development of rut depth.

Thoughts of brain EEG signal-to-text conversion using weighted feature fusion-based Multiscale Dilated Adaptive DenseNet with Attention Mechanism

Individuals with visual inefficiencies or different abilities face difficulties using their hands to operate smartphones and computers, necessitating reliance on others to enter data. Such dependence may lead to security and privacy issues, especially when sensitive information is shared with helpers. To address this problem, we present Think2Type, an efficient Brain-Computer Interface (BCI) that enables users to translate their active intentions into text format based on Morse code. BCI leverages brain activity to facilitate interaction with computers, often captured via Electroencephalography (EEG). This work proposes an enhanced attention-based deep learning strategy to develop an efficient text conversion mechanism from EEG signals. We begin by collecting EEG signals from standard benchmark datasets and extracting spectral and statistical features in phase 1, concatenating them into concatenated feature set 1 (F1). In phase 2, we extract spatial and temporal features via a One-Dimensional Convolutional Neural Network (1DCNN) and a Recurrent Neural Network (RNN), respectively, concatenating them into concatenated feature set 2 (F2). Weighted feature fusion is performed on concatenated features F1 and F2, with the hybrid optimization algorithm Eurasian Oystercatcher Wild Geese Migration Optimization (EOWGMO) optimizing the weight for improved fusion efficiency. The text conversion phase utilizes the Multiscale Dilated Adaptive DenseNet with Attention Mechanism (MDADenseNet-AM) to obtain the converted text information. The MDADenseNet-A's parameters are optimized to improve thought-to-text conversion performance. The developed model's performance is evaluated via experimental analysis and compared to conventional techniques, resulting in a higher accuracy value of 96.41%, facilitating appropriate text conversion.

Process Simulation and Performance Models for Enhanced Modular Refinery Operations in Nigeria

The study aimed at availability of petroleum products in Nigeria through the operation of enhanced modular refinery process due to inefficient conventional major refineries. The enhanced modular refinery converts or processes residue product from conventional modular refinery as feedstock to the hydrocracker reactor for viable and desired products such as liquefied petroleum gas, naphtha and diesel. Therefore, twenty Nigerian crude oil types were classified as sweet, light and medium crude oil, and these crude oil types were categorized based on their recovery volume at true boiling point of 370oC as Group A, Group B and Group C respectively. Thus, based on product output and equipment cost, a modular refinery with 29 trays was used in this study, as light and medium sweet crude oil types were simulated in a modular refinery of various column trays prior to the desired tray. A topping plant with a 30,000 barrel per day capacity and a modified topping plant with 29 trays respectively were used to process different types of Nigerian crude oil. using Aspen Hysys to evaluate their products yield and tray compositions. The modified modular refinery with hydrocracker yielded more valuable from the residue of conventional modular refinery with minimal bottom fraction. Performance models for hydrocracker reactor was developed based on the nature of reaction, kinetic parameters estimated and results compared with experimental data with minimum deviations. The developed performance models predicted feedstock conversion and product yield along the hydrocracker reactor’s dimensionless length by solving a set of ordinary differential equation models. Thus, the hydrocracking process was simulated to evaluate the effects of catalyst effectiveness factor on feedstock conversion and products yield.

Synthesis and efficacy of cactus-banana peels composite as a natural coagulant for water treatment

Aluminium and ferric salts continue to be used as coagulants in drinking water treatment. Natural coagulants can be used for the same purpose because they are cheaper, locally accessible and environmentally friendly. However, low production yields and high operation costs affect commercial adoption of natural coagulants from individual plants, hence the exploration of performance of their composites. This study evaluated the performance of cactus-banana peels composite as natural coagulant for water treatment because of the low cost nature of the two plants. Design Expert Software was used to design jar test experiments for attainment of optimum mixing ratio of the composite for determining optimum dosage, pH and extraction time for development of performance models. Performance of the coagulant was evaluated based on removal efficiencies of turbidity, total suspended solids (TSS) and Escherichia coli (E.coli). The goodness of fit for developed models was evaluated using R2 values and adequate precision. The optimal composition of the composite was cactus to banana peels ratio 0.62:0.38. The optimally mixed powder had a bulk density of 590 kg/m3 while the extracted liquid coagulant had pH and electrical conductivity of 7.05 and 1123 μs/cm, respectively. The optimum dosage, pH and extraction time were 12.25 ml/l, 7.31 and 26.53 min, respectively. Turbidity, TSS and E. coli removal efficiencies were 87.13, 82.15 and 84.02%, respectively. These results indicated good performance of the composite coagulant in water treatment compared to 82–99% for alum, the most commonly used commercial coagulant.

Regional implementation of the mechanistic empirical pavement design and analysis approach: Egyptian case study

The Mechanistic-Empirical (ME) pavement design approaches have many advantages including; considering a wide range of traffic loadings along with a diversity of climatic conditions. This is in addition to the ability to comprehensively and thoroughly characterizing the pavement materials, which consequently led to the enhanced capability of predicting pavement performance. However, the implementation of the ME procedures has many challenges including the huge amount of traffic and climatic data needed, the needed advanced characterization of the pavement materials, and the need for the calibration and validation to the local conditions. Accordingly, an increased required time to develop and evaluate the design is warranted. Furthermore, the lack of the experience with the ME pavement design approaches is considered an additional challenge. In order to encourage the ME design procedures implementation in a scarcity of data and lack of experience regions, this research proposes a ME regional implementation model. This model was developed then implemented in Egypt as a case study representing the developing countries of scarcity of both data and experience of the ME approaches. First, libraries of the local material/traffic inputs were created. Second, the performance of typical designed pavement sections was predicted using the ME approach in terms of the most regional prevailing performance indicators; rutting, fatigue cracking, and thermal cracking. The Machine Learning (ML) technique was then used to develop correlation models. Third, the developed models were presented in the form of ME pavement design/analysis tool. Three performance ML based-models were developed with good R2 (coefficient of determination) of 0.789, 0.755, and 0.959 for fatigue cracking, thermal cracking, and rutting, respectively. Finally, the developed performance models were implemented in the form of a user-friendly design/analysis tool ready for use as a mechanistic empirical tool developed based on the regional conditions. The Egyptian case study proved a good validation and verification of the developed model.

Profiling the BLAST bioinformatics application for load balancing on high-performance computing clusters

BackgroundThe Basic Local Alignment Search Tool (BLAST) is a suite of commonly used algorithms for identifying matches between biological sequences. The user supplies a database file and query file of sequences for BLAST to find identical sequences between the two. The typical millions of database and query sequences make BLAST computationally challenging but also well suited for parallelization on high-performance computing clusters. The efficacy of parallelization depends on the data partitioning, where the optimal data partitioning relies on an accurate performance model. In previous studies, a BLAST job was sped up by 27 times by partitioning the database and query among thousands of processor nodes. However, the optimality of the partitioning method was not studied. Unlike BLAST performance models proposed in the literature that usually have problem size and hardware configuration as the only variables, the execution time of a BLAST job is a function of database size, query size, and hardware capability. In this work, the nucleotide BLAST application BLASTN was profiled using three methods: shell-level profiling with the Unix “time” command, code-level profiling with the built-in “profiler” module, and system-level profiling with the Unix “gprof” program. The runtimes were measured for six node types, using six different database files and 15 query files, on a heterogeneous HPC cluster with 500+ nodes. The empirical measurement data were fitted with quadratic functions to develop performance models that were used to guide the data parallelization for BLASTN jobs.ResultsProfiling results showed that BLASTN contains more than 34,500 different functions, but a single function, RunMTBySplitDB, takes 99.12% of the total runtime. Among its 53 child functions, five core functions were identified to make up 92.12% of the overall BLASTN runtime. Based on the performance models, static load balancing algorithms can be applied to the BLASTN input data to minimize the runtime of the longest job on an HPC cluster. Four test cases being run on homogeneous and heterogeneous clusters were tested. Experiment results showed that the runtime can be reduced by 81% on a homogeneous cluster and by 20% on a heterogeneous cluster by re-distributing the workload.DiscussionOptimal data partitioning can improve BLASTN’s overall runtime 5.4-fold in comparison with dividing the database and query into the same number of fragments. The proposed methodology can be used in the other applications in the BLAST+ suite or any other application as long as source code is available.

Simulation of packed bed reactors for naphtha reforming unit

The study focused on the performance evaluation of catalytic naphtha reforming unit via a packed bed reactors arranged in series. Three series arranged packed bed naphtha reforming reactors were developed and simulated using Aspen hysys software and steady state performance models of the packed bed naphtha reforming reactor were developed through the application of the principle of conservation of mass and energy to predict both the yields of theses reactors, their dimensions in terms of length, diameter, height, volume and temperature effects or progressions along the reactors’ length respectively. Three main reaction paths were considered in this research study that include the conversion of naphthenes to aromatics, naphthenes to paraffins and hydrocracking of paraffins. The developed performance models yielded first order ordinary differential equations which were solved using Runge-Kutta fourth order ODE45 solver. The packed bed reactor operating conditions of temperature and pressure are between 4800C and 5200C and 2atm respectively with ninety percent (90%) conversion. The simulated reactor performance model for the three reactors yielded 29.7 m3, 40.78 m3 and 52.33 m3 for reactor volume, catalyst bed height of 11m, 12.98 m, 16 m, and reactor’s diameter of 1.85 m, 2 m and 2.04 m respectively.

Performance and power modeling and prediction using MuMMI and 10 machine learning methods

SummaryEnergy‐efficient scientific applications require insight into how high performance computing system features impact the applications' power and performance. This insight can result from the development of performance and power models. In this article, we use the modeling and prediction tool MuMMI (Multiple Metrics Modeling Infrastructure) and 10 machine learning methods to model and predict performance and power consumption and compare their prediction error rates. We use an algorithm‐based fault‐tolerant linear algebra code and a multilevel checkpointing fault‐tolerant heat distribution code to conduct our modeling and prediction study on the Cray XC40 Theta and IBM BG/Q Mira at Argonne National Laboratory and the Intel Haswell cluster Shepard at Sandia National Laboratories. Our experimental results show that the prediction error rates in performance and power using MuMMI are less than 10% for most cases. By utilizing the models for runtime, node power, CPU power, and memory power, we identify the most significant performance counters for potential application optimizations, and we predict theoretical outcomes of the optimizations. Based on two collected datasets, we analyze and compare the prediction accuracy in performance and power consumption using MuMMI and 10 machine learning methods.

A visual analytics approach for the assessment of information quality of performance models—a software review

In this paper we provide a review of the main functionalities of a Visual Analytics Environment (VAE) developed for the assessment of data and information quality in the context of performance evaluation of research organizations. Performing data and information quality tests are necessary procedures to ensure the bibliometric and research performance evaluation analysis of organizations have the necessary robustness. The proposed environment is helpful to guide the user to an Information Quality-aware development of Performance models. This interactive visual analytics environment offers to the user the possibility to produce and compare information quality-aware indicators, exploring and defining correct behavior, identifying anomalous cases from both data quality and information quality perspectives, and supporting the user in forming hypotheses on possible causes for those anomalies. The proposed approach, exploiting visual interactive exploration results in a more efficient process, minimizing the number of cases for which a manual investigation is needed. The illustration on European higher education institutions data demonstrates the use of the presented functionalities and their benefits.

Investigation of fault detection and isolation accuracy of different Machine learning techniques with different data processing methods for gas turbine

Classification is an essential task for many applications, including text classification, image classification, data classification, and so on. The present study investigates the accuracy of different machine learning classification algorithms with three different data smoothing techniques for gas turbine fault detection and isolation task. The gas turbine performance model was developed by considering variable inlet guide vane and bleed air. Fouling and erosion were injected into all six main components of the gas turbine engine. Faulty and non-faulty data were generated from the developed performance model. Based on sensitivity analysis, 12 measurement parameters and 11,824 data points were selected for the development of a fault detection and isolation model. The faulty and non-faulty data were balanced, smoothed, corrected and normalized. Finally, the classification accuracy of the machine learning techniques was analyzed. The result shows that K-Nearest Neighbours, Neural Network and Decision Tree classifiers exhibited high classification accuracy, about 99% with all three data smoothing techniques. It is also observed that the computation time of Support Vector Machine is higher whereas K-Nearest Neighbours shows the lowest. Finally, the research proves that K-Nearest Neighbours is the best classification technique for gas turbine engine fault detection and isolation application.

Validation of a lithium-ion commercial battery pack model using experimental data for stationary energy management application.

Background:A cost-effective solution for the design of distributed energy storage systems implies the development of battery performance models yielding a suitable representation of its dynamic behaviour under realistic operation conditions. Methods:In this work, a lithium-ion battery (LIB) is tested to be further modelled and integrated into an existing energy management control system. This specific LIB (5.0 kW /9.8 kWh) is integrated with a commercial inverter and solar photovoltaic (PV) system (3.3 kWp) as part of a microgrid that is also encompassing other energy storage technologies at the University of Évora, Pole of INIESC - National Research Infrastructure for Solar Energy Concentration. A testing protocol fully characterizes the battery and the inverter efficiency to describe their performance better. Then, a battery model is built upon both the existent LIB description and experimental fitting regression. The model allows obtaining the voltage curve, the internal resistance (i.e., to describe instantaneous voltage drop/rise and transients), and the state of charge (SOC) and/or energy capacity based on the current input. The developed model is validated through the comparison with the experimental results. Results:The model approach presented a higher voltage RMSE (root mean square error) of 5.51 V and an MRE (maximum relative error) of 5.68 % in the discharge state. Regarding SOC, the MRE obtained was approximately 6.82 %. In the charge state, the highest RMSE voltage was 5.27 V, with an MRE of 6.74 %. Concerning SOC, the MRE obtained was approximately 6.53 %. Conclusions:The developed model is validated through the comparison with experimental results. Based on computational effort, simplicity of use and the associated model error, the approach is validated to the regular conditions of the commercial battery pack to be incorporated in the next research step, following a bottom-up modelling approach for an increasingly more complex smart grid.

The effect of customer satisfaction on parcel delivery operations using autonomous vehicles: An agent-based simulation study

The quality of Third-Party Logistics (3PL) services represented by delivery time decides the outcome of customer satisfaction. The result of this satisfaction judges the type of Word of Mouth (WoM) that, if positive, plays a vital role in attracting non-customers who are willing in 3PL services to join as customers.In this paper, we investigate the effect of an essential factor represented by Word of Mouth on the number of customers in 3PL companies. Therefore, an agent-based model for parcel delivery is developed to investigate the impact of social factors such as WoM and other operational factors, including vehicle number and speed, on customer number and satisfaction, average service time, and vehicle utilization.As a methodology, state charts of Vehicle, Customer, Hub agents are developed to mimic the messaging protocols between these agents under the WoM concept. A case study based in 3PL in Jordan is used as a test bench of the developed model. A sensitivity analysis study is conducted to test the developed model's performance, including different levels of influential model parameters such as targeting non-customers parameters by Loyal/Unhappy customers.Key results reveal that the best scenario is achieved when the WoM value equals 10, the vehicle number equals 30, and the vehicle speed equals 60 km/h. These model parameters result in higher customer numbers of 873, vehicle utilization equals 63%, and customer satisfaction equals 99%. Video of our proposed model showing it in action can be found at: https://www.youtube.com/watch?v=3rR4l130-QU.

Developing Network Slurry Seal Performance Models Using Pavement Management Systems Data

As a popular pavement preservation treatment, a slurry seal is typically included in agencies’ maintenance and rehabilitation toolboxes, in decision trees of pavement management systems (PMS), or both. However, performance prediction models for slurry seals are mostly anecdotal or at undertaken at project level. The goal of this paper is to illustrate a method of developing data-driven performance models for slurry seal applications for use at a network level. A total of 537,891 pavement condition records were collected from 18 different agencies in the United States using the Metropolitan Transportation Commission’s PMS program, StreetSaver®. Of these records, 1,195 slurry seal projects met the requirements to be utilized in the development of slurry seal performance models. The development of slurry seal performance models includes grouping project-level modeling results into functional classes, and testing and validating the slurry seal performance models with test cases as well as on actual data produced by agencies. The performance models for slurry seals placed on an asphalt concrete pavement vary depending mainly on the existing pavement’s performance, the timing of the slurry seal application, the functional class, and surface type. The data-driven slurry seal models improved the accuracy of predicting slurry seal performance using PMS. The models could also be used to develop beneficial area curves, which are useful in estimating the cost-effectiveness of slurry seals constructed at various pavement condition index values and in identifying the optimal timings for slurry seal applications at a network level.

Validation of a lithium-ion commercial battery pack model using experimental data for stationary energy management application

Background:A cost-effective solution for the design of distributed energy storage systems implies the development of battery performance models yielding a suitable representation of its dynamic behaviour under realistic operation conditions.Methods:In this work, a lithium-ion battery (LIB) is tested to be further modelled and integrated into an existing energy management control system. This specific LIB (5.0 kW /9.8 kWh) is integrated with a commercial inverter and solar photovoltaic (PV) system (3.3 kWp) as part of a microgrid that is also encompassing other storage technologies at the University of Évora, Pole of INIESC – National Research Infrastructure for Solar Energy Concentration. The battery and the inverter are fully characterized through the implementation of a testing protocol aiming at better describing the battery performance. Then, a battery model is built upon both the existent LIB description and experimental fitting regression, for real-time predictive optimization control development. Considering the pre-determined efficiency of the inverter, the model allows to obtain the voltage curve, the series resistance (i.e., to describe instantaneous voltage drop/rise and transients), and the state of charge (SOC) and/or energy capacity, based on the current input. The developed model is validated through the comparison with the experimental results.Results:In discharge state, the model approach presented a higher voltage RMSE (root mean square error) of 5.51 V and an MRE (maximum relative error) of 5.68 %. Regarding SOC the MRE obtained was approximately 6.82 %. In charge state, the highest RMSE voltage was 5.27 V, with an MRE of 6.74 %. Concerning SOC, the MRE obtained was approximately 6.53 %. Conclusions:The developed setup allowed us to perform the necessary characterization tests under real operating conditions. Based on computational effort, simplicity of use, and the associated model error compared with the experimental data, generally, the model describes the battery behaviour.

Success determinants of projects of business software implementation: Research framework

Abstract The experience of many organizations so far has shown that the risk of failure of the software implementation project is high, and in order to reduce this risk, significant efforts have been made so far to implement the project management methodology. However, research to date shows that, despite improvements in project management, a relatively significant number of projects still fail by some of the performance criteria. A methodology is proposed that includes: (i) analysis of previous research; (ii) design a research instrument to measure project performance and define key determinants of project performance; (iii) development of a multi-criteria model for measuring the performance of software projects; (iv) development of software project management performance models for Croatian organizations using a data mining approach. These activities can be carried out using the following methods: bibliographic analysis of previous research, survey research method, descriptive and inferential statistics methods, structural equations method and data mining methods (supervised and unsupervised learning, text mining). The results of this research framework can be used as a starting point for researching the success of business software implementation projects.