Table 1: Comparison of different track deterioration models.

Approach		Variables	Strengths	Weaknesses
Mechanistic		• Track settlement, • Track deformation, • Track geometry (e.g. gauge), • Track Quality Index (TQI).	• Based on laboratory experiment data sources, • Clearly address track settlement and degradation, • Suitable for maintenance of a particular section of rail track.	• Challenging, intensive, time consuming. • Measurement of the affecting variables of rail structure may be difficult or poorly understood. • Materials of rail structure are not homogenous. • Difficulties in applying the model for different sections of rail track.
Statistical (Empirical)	Deterministic	• Traffic volume, • Dynamic axle, • Speed, • Accumulated tonnage (MGT), • Axle loads.	• Work well for large data sets.	• Potential to miss important degradation factors during application, • It does not account for uncertainty (i.e. input parameters and model geometry are not well known).
	Probabilistic	• Speed restrictions or line closure, • Track Quality Index (TQI), • Standard deviation of longitudinal level defects (SDLL) and horizontal alignment defects (SDHA), • Number of cracks missed by USI per year, • Rail breakage.	• Reasonable procedure and realistic findings, • Ability to deal with large numbers of datasets to achieve more accurate results.	• Not common due to lack of historical data, • Difficulties in predicting probability of track deterioration, • Bayesian models rely on Markov models especially when high numerical dimensions occur.
	Stochastic	• Time, • Degradation rate of longitudinal level.	• Ability to deal with large numbers of datasets to achieve more accurate results,	• No evidence to validate the claim of an exponential deterioration pattern.
Mechanical-empirical		• Track Quality Index (TQI), • Traffic parameters, • Maintenance parameters (EMGT), • Degradation Coefficient [58], • Time.	• Applicable to different track segments (e.g. curves, turnouts, straight lines), • Applicable to more accurate and less costly future maintenance procedures.	• Showing a higher rate of deterioration of lines in bridges, curve-bridges and turnouts in comparison with other model types.
Artificial Intelligence	Artificial Neural Networks (ANNs)	• Number of layers, • Nodes, • Type of the network and functions.	• Calibrating model with an optimization algorithm, • Optimising parameters of model.	• Presence of many effective factors resulting in more errors, • Validation of membership functions.
	Neuro-Fuzzy	• Fuzzy sets, • Fuzzy membership functions.	• Finding fuzzy rules from numerical data, • Considering human imprecise perception, • Categorising variables into different categories	• Complexity in abstracting fuzzy rules, • Connections of a proposition may be imprecise, • Difficulty in calibrating model parameters.