Research article

Decision trees and multi-level ensemble classifiers for neurological diagnostics

  • Received: 10 December 2013 Accepted: 23 June 2014 Published: 30 June 2014
  • Cardiac autonomic neuropathy (CAN) is a well known complication of diabetes leading to impaired regulation of blood pressure and heart rate, and increases the risk of cardiac associated mortality of diabetes patients. The neurological diagnostics of CAN progression is an important problem that is being actively investigated. This paper uses data collected as part of a large and unique Diabetes Screening Complications Research Initiative (DiScRi) in Australia with data from numerous tests related to diabetes to classify CAN progression. The present paper is devoted to recent experimental investigations of the effectiveness of applications of decision trees, ensemble classifiers and multi-level ensemble classifiers for neurological diagnostics of CAN. We present the results of experiments comparing the effectiveness of ADTree, J48, NBTree, RandomTree, REPTree and SimpleCart decision tree classifiers. Our results show that SimpleCart was the most effective for the DiScRi data set in classifying CAN. We also investigated and compared the effectiveness of AdaBoost, Bagging, MultiBoost, Stacking, Decorate, Dagging, and Grading, based on Ripple Down Rules as examples of ensemble classifiers. Further, we investigated the effectiveness of these ensemble methods as a function of the base classifiers, and determined that Random Forest performed best as a base classifier, and AdaBoost, Bagging and Decorate achieved the best outcomes as meta-classifiers in this setting. Finally, we investigated the meta-classifiers that performed best in their ability to enhance the performance further within the framework of a multi-level classification paradigm. Experimental results show that the multi-level paradigm performed best when Bagging and Decorate were combined in the construction of a multi-level ensemble classifier.

    Citation: Herbert F. Jelinek, Jemal H. Abawajy, Andrei V. Kelarev, Morshed U. Chowdhury, Andrew Stranieri. Decision trees and multi-level ensemble classifiers for neurological diagnostics[J]. AIMS Medical Science, 2014, 1(1): 1-12. doi: 10.3934/medsci.2014.1.1

    Related Papers:

    [1] Gözde Özkan Tükel . Integrable dynamics and geometric conservation laws of hyperelastic strips. AIMS Mathematics, 2024, 9(9): 24372-24384. doi: 10.3934/math.20241186
    [2] Tamara M. Garrido, Rafael de la Rosa, Elena Recio, Almudena P. Márquez . Conservation laws and symmetry analysis of a generalized Drinfeld-Sokolov system. AIMS Mathematics, 2023, 8(12): 28628-28645. doi: 10.3934/math.20231465
    [3] Xue Geng, Liang Guan, Dianlou Du . Action-angle variables for the Lie-Poisson Hamiltonian systems associated with the three-wave resonant interaction system. AIMS Mathematics, 2022, 7(6): 9989-10008. doi: 10.3934/math.2022557
    [4] Gulgassyl Nugmanova, Aidana Azhikhan, Ratbay Myrzakulov, Akbota Myrzakul . Nurshuak-Tolkynay-Myrzakulov system: integrability, geometry and solutions. AIMS Mathematics, 2025, 10(6): 14167-14182. doi: 10.3934/math.2025638
    [5] Mohammad Nazrul Islam Khan, Uday Chand De . Liftings of metallic structures to tangent bundles of order $ r $. AIMS Mathematics, 2022, 7(5): 7888-7897. doi: 10.3934/math.2022441
    [6] Yanhui Bi, Zhixiong Chen, Zhuo Chen, Maosong Xiang . The geometric constraints on Filippov algebroids. AIMS Mathematics, 2024, 9(5): 11007-11023. doi: 10.3934/math.2024539
    [7] George Rosensteel . Differential geometry of collective models. AIMS Mathematics, 2019, 4(2): 215-230. doi: 10.3934/math.2019.2.215
    [8] Mingliang Zheng . Study on the symmetries and conserved quantities of flexible mechanical multibody dynamics. AIMS Mathematics, 2023, 8(11): 27969-27982. doi: 10.3934/math.20231430
    [9] Huizhang Yang, Wei Liu, Yunmei Zhao . Lie symmetry reductions and exact solutions to a generalized two-component Hunter-Saxton system. AIMS Mathematics, 2021, 6(2): 1087-1100. doi: 10.3934/math.2021065
    [10] Carey Caginalp . Minimization solutions to conservation laws with non-smooth and non-strictly convex flux. AIMS Mathematics, 2018, 3(1): 96-130. doi: 10.3934/Math.2018.1.96
  • Cardiac autonomic neuropathy (CAN) is a well known complication of diabetes leading to impaired regulation of blood pressure and heart rate, and increases the risk of cardiac associated mortality of diabetes patients. The neurological diagnostics of CAN progression is an important problem that is being actively investigated. This paper uses data collected as part of a large and unique Diabetes Screening Complications Research Initiative (DiScRi) in Australia with data from numerous tests related to diabetes to classify CAN progression. The present paper is devoted to recent experimental investigations of the effectiveness of applications of decision trees, ensemble classifiers and multi-level ensemble classifiers for neurological diagnostics of CAN. We present the results of experiments comparing the effectiveness of ADTree, J48, NBTree, RandomTree, REPTree and SimpleCart decision tree classifiers. Our results show that SimpleCart was the most effective for the DiScRi data set in classifying CAN. We also investigated and compared the effectiveness of AdaBoost, Bagging, MultiBoost, Stacking, Decorate, Dagging, and Grading, based on Ripple Down Rules as examples of ensemble classifiers. Further, we investigated the effectiveness of these ensemble methods as a function of the base classifiers, and determined that Random Forest performed best as a base classifier, and AdaBoost, Bagging and Decorate achieved the best outcomes as meta-classifiers in this setting. Finally, we investigated the meta-classifiers that performed best in their ability to enhance the performance further within the framework of a multi-level classification paradigm. Experimental results show that the multi-level paradigm performed best when Bagging and Decorate were combined in the construction of a multi-level ensemble classifier.


    1. Introduction

    When a nonlinear evolution equation can be generated from a pair of linear partial differential equations of first order by means of a compatibility condition, the nonlinear evolution equation is said to be Lax integrable and the nonlinear system is called a Lax pair [7,8]. However, if such a pair of equations can be determined, then such objects such as gauge and Darboux transformations as well as an infinite number of conservation laws can be constructed for the equation. Complete integrability is another property associated with such systems. This concept is usually formulated for a distribution on a manifold [9].

    Definition 1. $(i)$ Let $m$, $n$ be integers with $1 \leq m \leq n$, then an $m$-dimensional distribution $\mathcal{D}$ on an $n$-dimensional manifold $M$ is a selection of an $m$-dimensional subspace $\mathcal{D}_p$ of $T_p M$ for each $p \in M$. The distribution is $C^{\infty}$ if for $p \in M$, there exists a neighborhood $U$ of $p$ and $m$ vector fields $X_1, \ldots, X_m$ on $U$ which span $\mathcal{D}$ at each point in $U$. A vector field is said to lie in the distribution $\mathcal{D}$ if $X_p \in \mathcal{D}_p$, for each $p \in M$. $(ii)$ A $C^{\infty}$ distribution is called involutive, or completely integrable, if $[X, Y] \in \mathcal{D}$ whenever $X, Y \in \mathcal{D}$. $(iii)$ We say $\mathcal{D}$ is integrable if for any $p \in M$ there is a (local) submanifold $N \subset M$ called a (local) integral manifold of $\mathcal{D}$ at $p$ containing $p$ whose tangent bundle is exactly $\mathcal{D}$ restricted to $N$.

    An equivalent way of stating this is that a distribution $\mathcal{D}$ is completely integrable if through each point $p \in M$, there exists an integral manifold whose dimension is equal to the dimension of $\mathcal{D}$. The Frobenius theorem gives the necessary and sufficient condition for the integrability of a distribution. The theorem can be expressed in several different but equivalent forms, in terms of vector fields as well as differential forms.

    The exterior differential version of the Frobenius theorem states that a distribution $\mathcal{D}$ is completely integrable if and only if the ideal $I (\mathcal{D})$ generated by $\mathcal{D}$ is a closed differential ideal, hence $d \, I (\mathcal{D}) \subset I (\mathcal{D})$. This is a very practical form of the theorem.

    The objective here is to investigate $1+1$ dimensional nonlinear evolution equations which possess Lax integrability by means of a geometric approach. It will be seen that the relation between complete and Lax integrability can be formulated in this way when the Lax integrability of the nonlinear evolution equation is given in terms of exterior differential forms [11]. This permits geometric interpretations of the Lax equation to be given. Exterior differential systems also provides an efficient way to formulate the concept of gauge and Darboux transformations for such equations. The terminology nonlinear evolution equation will be understood to signify an equation in one space and one time variable everywhere, that is, in one$+$one dimensions[2,3,4].


    2. Lax Representations in Two Independent Variables

    Historically in the theory of nonlinear evolution equations of soliton type, it is well known that these equations may be considered as sufficient conditions for the integrability of an eigenvalue problem involving a set of linear partial differential equations which contain the solution of the corresponding evolution equation. As a particular example [5] the integrability of the linear system

    $ y_x = \lambda + u, \qquad y_t = - u_{xx} + \frac{6}{n+1} (-u)^{n+1} + \mu, $

    leads to the nonlinear system

    $ u_t + u_{xxx} + 6 (-u)^n \, u_x =0. $

    If we eliminate $u$ from the pair, it is found that $y= y(x, t)$ satisfies the equation

    $ y_t + y_{xxx} - \frac{6}{n+1} ( \lambda - y_x)^{n+1} - \mu=0. $

    This is often called the potential equation when $\lambda=\mu=0$, and is an example of a Bäcklund transformation as well

    The Lax representation of a nonlinear evolution equation in two independent variables considered here is introduced as a matrix system,

    $ \psi_x = M (\lambda) \psi, \qquad \psi_t = N ( \lambda) \psi. $ (2.1)

    The parameter $\lambda$ in (2.1) is usually called a spectral parameter, $\psi$ is referred to as an eigenfunction associated with the spectral parameter $\lambda$, and $M$, $N$ are particular matrices whose elements depend on $\lambda$. The sizes of the matrices may be left arbitrary for the moment. Collectively (2.1) is referred to as a Lax pair. The nonlinear equation which corresponds to $M$ and $N$ is obtained by differentiating the first equation with respect to $t$ and the second with respect to $x$ and equating the mixed partials of $\psi$ to produce the compatibility condition for (2.1)

    $M_t - N_x + [M, N] = 0. $ (2.2)

    The bracket in (2.2) is defined as $[M, N] = MN -NM$.

    To obtain a geometric picture of this type of integrability, define a matrix of one-forms $\omega$ as

    $ \omega = M \, dx + N \, dt. $ (2.3)

    System (2.1) results as a consequence of the following relation

    $ \sigma = d \psi - \psi \omega =0. $ (2.4)

    Exterior differentiation of (2.3) produces the result,

    $ d \sigma = - d \omega \, \psi + \omega \wedge \, d \psi =0. $ (2.5)

    Solving (2.4) for $d \psi$ and putting this in (2.5), we obtain,

    $ d \sigma = - d \omega \, \psi + \omega \wedge ( \sigma + \omega \, \psi ) = \omega \wedge \sigma - ( d \omega - \omega \wedge \omega ) \, \psi. $ (2.6)

    The Frobenius theorem implies system (2.3) is completely integrable if and only if the system of one-forms $\omega$ satisfies the condition

    $ \Omega = d \omega - \omega \wedge \omega =0. $ (2.7)

    In (2.7), $\Omega$ is a square matrix of two-forms usually referred to as the curvature. It is straightforward to verify that (2.7) is equivalent to (2.2). These equations, (2.2) and (2.7), are referred to as zero curvature conditions, and (2.7) is equivalent under (2.3) to the nonlinear evolution equation whose Lax representation is (2.1).

    Theorem 1. The nonlinear evolution equation in $(x, t)$ whose Lax representation is given by (2.1) must be completely integrable.

    Proof: Differentiating the curvature $\Omega$ in (2.7) gives

    $ d \Omega = - d \omega \wedge \omega + \omega \wedge d \omega =- ( \Omega + \omega \wedge \omega) \wedge \omega + \omega \wedge ( \Omega + \omega \wedge \omega ) $
    $ =- \Omega \wedge \omega + \omega \wedge \Omega. $ (2.8)

    Therefore, system (2.7) is completely integrable according to the Frobenius theorem. Hence (2.7) corresponds to the nonlinear evolution equation whose Lax representation is (2.1) as required.

    There is an equivalent approach to the definition of Lax integrability which is discussed now. Let $U \subset \mathbb R^2$ be coordinatized by $x$ and $t$ and let $V = U \times \mathbb R^n \subset \mathbb R^{n+2}$ be a fiber bundle with $U$ as the base manifold and $U$ carries coordinates related to a particular nonlinear evolution equation.

    Suppose the equation can be represented by a system of two-forms defined on $V$. As a specific example, consider the system of two-forms

    $ \alpha_1 = d u \wedge dt - p \, dx \wedge dt, $
    $ \alpha_2 = dp \wedge dt - q \, dx \wedge dt, $ (2.9)
    $ \alpha_3 =-du \wedge dx + dq \wedge dx - u \, dq \wedge dt + u \, du \wedge dt + \beta \, (u-q) \, du \wedge dt. $

    The parameter $\beta$ which appears is a constant. If the map $s : U \to V$ is a cross section of $V$ with the property

    $ s^* \, \alpha_i =0, $ (2.10)

    where the $\alpha_i$ are a system of forms such as(2.9), and mapping $s^*$ denotes the pull-back of $s$, it can be verified that $u (x, t)$ is a solution of the associated equation. Conversely, for any given solution $u (x, t)$ of the equation, the map $s : U \to V$ which is given by $s (x, t) = (x, t, u (x, t), u_x (x, t), \cdots)$ is a cross section of $V$ satisfying $s^* \alpha_i =0$.

    Differentiating the forms in (2.9) gives

    $ d \alpha_1 =- dp \wedge dx \wedge dt = dx \wedge \alpha_2, $
    $ d \alpha_2 = dx \wedge ( - \alpha_3 + u (( 1 + \beta)u -q ) \, \alpha_1), $
    $ d \alpha_3 = ( 1 - \beta) [d q \wedge \alpha_1 + p \, dt \wedge \alpha_3 -p \, dx \wedge \alpha_1]. $

    Clearly, all of the $d \alpha_j$ vanish modulo (2.9) so the $\{ \alpha_i \}$ represent a closed differential ideal. Sectioning $\alpha_1$ and $\alpha_2$ gives $p=u_x$ and $q = u_{xx}$ and the nonlinear equation results from evaluating

    $ 0= \alpha_3 |_s = ( u_t - q_t + u (u_x - q_x) + \beta ( u-q) u_x ) \, dx \wedge dt. $

    These results imply the partial differential equation

    $ (u-q)_t +u (u-q)_x + \beta (u-q) u_x =0. $

    Introducing $\rho = u -u_{xx}$, then for the case $\beta=3$, the Degasperis-Procesi equation results,

    $ \rho_t + \rho_x u + 3 \rho u_x =0, $

    and for $\beta =2$, the Camassa-Holm equation appears

    $ \rho_t + \rho_x u + 2 \rho \, u_x =0. $

    It is important to note that if a set of forms $\{ \alpha_i \}$ can be determined which correspond to a particular evolution equation these forms can be used to construct both Lax pairs and Bäcklund transformations for that equation.

    For any nonlinear evolution equation, a corresponding set of two forms can be constructed. In fact, two systems of two-forms $\{ \alpha_i \}$ and $\{ \beta_i \}$ are said to be equivalent if one set can be expressed in terms of the other as $\beta_i = f^j_i \, \alpha_j$, and the rank of the matrix $(f^j_i)$ is maximal.

    Theorem 2. (Lax Integrability) A nonlinear evolution equation in $(x, t)$ is Lax integrable if and only if there exists a square matrix $\omega$ of one-forms in $(dx, dt)$ such that the system (2.7) is equivalent to $\{ \alpha_i =0 \}$.

    Proof: Assume the Lax representation for the nonlinear equation is given by (2.1). If $\omega$ is defined to be the one-form (2.3), then $\omega$ is exactly what is required according to the demonstrated equivalence of relations starting with (2.7) implying (2.2) which then yields the nonlinear evolution equation (2.1) which finally can be put in the equivalent differential form $\{ \alpha_i =0 \}$.

    Second suppose the matrix of one-forms is $\omega$, then $\omega$ can be identified as a combination of a $dx$ and $dt$ term, so (2.3) holds. Then $(M, N)$ can be identified as the Lax pair demanded by $\Omega=0$.


    3. Prolongation and Differential Systems

    The Prolongation Method emerges in a natural way out of this theory. Prolongations are important for constructing solutions of evolution equations and also for deriving Bäcklund and auto-Bäcklund transformations which transform a solution of an equation into a solution of another or the same equation. The main ideas seem to originate with Wahlquist and Estabrook [10,12,13]. The nonlinear equation with Lax representation given by (2.1) can be replaced by an equivalent system of two-forms $\{ \alpha_i =0 \}$ defined on the manifold $V$. Now introduce an additional system of Pfaff forms $\varphi_i$ on the vector bundle $E = V \times \mathbb R^n$ which are defined as follows

    $ \varphi_i = d y_i - F_i \, dx - G_i \, dt. \qquad i =1, \ldots, n, $ (3.1)

    The new variables $y_i$ are often called pseudopotentials and provide a coordinate system for $\mathbb R^n$. The $\varphi_i$ are one-forms on $E$ and the $F_i$, $G_i$ are a set of functions which may depend on all the coordinates of $E$. The prolongation method requires the ideal which is generated by the system $\{ \alpha_i =0, \varphi_i =0 \}$ to be a closed differential ideal. This will provide constraints which serve to determine the form of the unknown functions $F_i$ and $G_i$ in (3.1). Thus, the prolongation condition has to have the following structure: the derivative of (3.1) must be of the following form,

    $ d \varphi_i = \beta^j_i \, \alpha_j + \gamma^j_i \wedge \varphi_j, \qquad i= 1, \ldots, n, $ (3.2)

    where $(\beta^j_i)$ is a matrix of functions and $(\gamma^j_i)$ is a matrix of one-forms. Formally comparing $\sigma$ with $\varphi_i$, the pseudo-potentials $y_i$ can be identified with the eigenfunction $\psi$, while $F_i$, $G_i$ correspond to $\omega$. There is also a formal correspondence between (3.2) and (2.6). Out of this correspondence, it may be seen that the prolongation condition guarantees that (3.1) is completely integrable on the solution manifold $\{ \alpha_i =0 \}$ of the equation whose Lax representation is (2.1). The Lax system (2.1) for the equation can be obtained with the complete integrability condition, although Lax integrability is a stronger property than complete integrability.

    A geometric interpretation comes directly from differential geometry for the Lax equation (2.4) [1,6]. Assume there is a connection $\nabla$ which is defined on the vector bundle $E$, and the sections $e_1, \cdots, e_n$ form a frame of sections of $E$. By using the frame of sections ${\cal S} = (e_1, \cdots, e_n)^T$ and the connection, and $n \times n$ connection matrix $\omega$ can be defined by the formula

    $ \nabla {\cal S} = \omega \, {\cal S}. $ (3.3)

    Elements of the connection matrix $\omega$ has elements which depend on the coordinates of the manifold $V$. If a section $s = \eta^i e_i$ is a parallel section of $E$ with the $\eta^i$ functions on $V$, from $\nabla s =0$, the $\eta^i$ will satisfy the following equations

    $ d \eta^i + \eta^j \omega^i_j =0, \qquad i=1, \cdots, n. $ (3.4)

    If we set $\eta = (\eta^1, \cdots, \eta^n)$, then (3.4) can be put in matrix form as

    $ d \eta + \eta \, \omega =0. $ (3.5)

    The connection $\nabla$ on the vector bundle $E$ induces a connection $\nabla^*$ on the dual vector bundle $E^* = V \times (\mathbb R^n)^*$, where $(\mathbb R^n)^*$ denotes the dual space of $\mathbb R^n$. If a dual frame of sections ${\cal S}^* = (e^{1*}, \cdots, e^{n*})^T$ of $E^*$ is established, then $\langle e_i, e^{j*} \rangle = \delta^j_i$ with $\langle, \rangle$ the inner product in the vector bundles $E$ and $E^*$. On the dual space, the connection is written $\nabla^* {\cal S}^* =-\omega {\cal S}^*$, so the induced connection matrix on the dual $E^*$ is $-\omega$. If the section $ s^* = \theta_i e^{i*}$ is a parallel section of $E^*$ where the $\theta_i$ are functions on $V$ with $\nabla^* s^* =0$, then the $\theta_i$ satisfy the equation

    $ d \theta^i - \theta_j \omega^j_i =0, \qquad i=1, \cdots, n. $

    Defining $\theta = (\theta_1, \cdots, \theta_n)$, then this can be put in matrix form is

    $ d \theta - \omega \, \theta =0. $ (3.6)

    Equation (3.6) is exactly Lax equation (2.4), and this Lax equation may be thought of as the parallel section equation on the dual vector bundle $E^*$ with connection matrix equal to $-\omega = -M \, dx -N \, dt$. The eigenfunctions $\psi$ correspond to the vector formed by the coordinates of the parallel section of $E^*$ under the dual frame of sections of $S^*$. Then the equation for $\Omega$ represents the curvature matrix where $\omega$ plays the role of connection matrix. Pfaff system (2.3) is completely integrable then if and only if curvature matrix $\Omega$ vanishes. Also, if the zero curvature condition $\Omega =0$ is satisfied, there exists $n$ linearly independent parallel sections, or alternatively, the Lax equation (2.4) has $n$ linearly independent solutions.


    4. Lax Systems and Gauge Transformations in Terms of Differential Systems

    Consider two Lax systems defined by

    $ \Psi_x = M \Psi, \qquad \Psi_t = N \Psi, $ (4.1)
    $ \Phi_x = \tilde{M} \Phi, \qquad \Phi_t = \tilde{N} \Phi. $ (4.2)

    If a gauge transformation $\tau$ exists such that

    $ \Phi = \tau \, \Psi, $ (4.3)

    where $\tau$ is a matrix that transforms (4.1) into (4.2), a system of equations satisfied by $\tau$ can be obtained. Differentiating (4.3) with respect to $x$ gives

    $ \Phi_x = \tau_x \Psi + \tau \Psi_x = \tau_x \Psi + \tau \, M \Psi. $

    This must be equal to $\tilde{M} \Phi = \tilde{M} \tau \Psi$, and so isolating $\tau_x \Psi$

    $ \tau_x \Psi = ( \tilde{M} \tau - \tau M ) \Psi. $ (4.4)

    This yields an equation for $\tau_x$. Similarly, evaluating $\Psi_t$ gives

    $ \tau_t \Psi = ( \tilde{N} \tau - \tau N ) \Psi. $ (4.5)

    Therefore, the matrix $\tau$ satisfies the following system,

    $ \tau_x = \tilde{M} \tau - \tau M, \qquad \tau_t = \tilde{N} \tau - \tau N. $ (4.6)

    In fact, system (4.6) can be expressed in the form [10]

    $ \tilde{\omega} = \tau \omega \tau^{-1} + d \tau \, \tau^{-1} $ (4.7)

    upon noting that $\omega= M \, dx + N \, dt$ and $\tilde{\omega} = \tilde{M} \, dx + \tilde{N} \, dt$. Now (4.7) is just the transformation formula for the connection matrix under the transformation of the basis of sections, namely,

    $ \tilde{S} = \tau \, S. $ (4.8)

    The transformation equation for the curvature matrix is given by

    $ \tilde{\Omega} = \tau \, \Omega \, \tau^{-1}, $ (4.9)

    where $\Omega$ and $\tilde{\Omega}$ are the respective curvature forms. This can be summarized as the following statement.

    Theorem 3. A nonlinear evolution equation which is Lax integrable is equivalent to the system $\tau \Omega \tau^{-1} =0$ for some $n \times n$ invertible matrix $\tau$ whose elements depend on the coordinates of $V$.

    Proof: If an equation with Lax integrability is equivalent to system (4.9), then the Lax equation corresponding to that equation is $d \tilde{\psi} = \tilde{\omega} \, \tilde{\psi}$, with $\tilde{\omega}$ given by (4.7).

    Conversely, it has been noted that every nonlinear evolution equation with Lax integrability is equivalent to a representative equation under gauge transformation. Therefore, there is an invertible matrix $\tau$ such that the given equation is equivalent to the system $\tau \Omega \tau^{-1}=0$.

    System (4.7) is clearly equivalent to the following Pfaff system

    $ \Theta = d \tau - \tilde{\omega} \, \tau + \tau \, \omega =0. $ (4.10)

    Expanding out (4.10) more fully, it is

    $ \Theta = \tau_x \, dx + \tau_t \, dt - (\tilde{M} \, dx + \tilde{N} dt) \tau + \tau ( M \, dx + N \, dt )=0. $

    Theorem 4. Pfaff system (4.10) is completely integrable.

    Proof: Differentiating (4.10), it is found that

    $ d \Theta = - d \tilde{\omega} \tau + \tilde{\omega} \wedge d \tau + d \tau \wedge \omega + \tau \, d \omega $
    $ =- d \tilde{\omega} \tau + \tilde{\omega} \wedge ( \Theta + \tilde{\omega} \tau - \tau \omega) + ( \Theta + \tilde{\omega} \tau - \tau \omega ) \wedge \omega + \tau \, d \omega $
    $ =- d \tilde{\omega} \tau + \tilde{\omega} \wedge \Theta + \tilde{\omega} \wedge \tilde{\omega} \tau - \tilde{\omega} \wedge \tau \omega + \Theta \wedge \omega + \tilde{\omega} \tau \wedge \omega - \tau \omega \wedge \omega + \tau \, d \omega $
    $ =- ( d \tilde{\omega} - \tilde{\omega} \wedge \tilde{\omega}) \tau + \tau ( d \omega - \omega \wedge \omega) + \tilde{\omega} \wedge \Theta + \Theta \wedge \omega. $ (4.11)

    Given that $\Omega =0$ and $\tilde{\Omega} =0$, the Frobenius theorem implies Pfaff system (4.11) is completely integrable and moreover,

    $ d \Theta = ( \tilde{\omega} - \omega ) \wedge \Theta. $ (4.12)

    Theorem 4 says that every nonlinear evolution equation in one$+$one dimensions with Lax integrability is gauge equivalent to any other. For example, if the KdV equation is taken as a typical representative, assuming the $n$-dimensional Lax representation for it is $d \psi = \omega \, \psi$, the KdV equation is equivalent to $\Omega =0$.

    Theorem 5. The nonlinear equation in $(x, t)$ with Lax integrability is equivalent to the system $\tau \Omega \tau^{-1} =0$ for some $n \times n$ invertible matrix $\tau$ whose elements are functions on the manifold $V$.

    Proof: If a nonlinear equation with Lax integrability is equivalent to the system $\tau \Omega \tau^{-1} =0$, the Lax equation corresponding to that equation is $d \tilde{\psi} = \tilde{\omega} \, \tilde{\psi}$, where $\tilde{\omega}$ is given by (4.7). Then that nonlinear evolution equation is Lax integrable.

    Conversely, it is known that every nonlinear evolution equation in one$+$one dimensions with Lax integrability is gauge equivalent to one of the equations, say the representative equation. Thus there is an invertible matrix $\tau$ such that the given nonlinear equation with Lax integrability is equivalent to the system $\tau \Omega \tau^{-1} =0$.

    If two systems $\Omega=0$ and $\tau \Omega \tau^{-1}=0$ correspond to the same nonlinear evolution equation with Lax integrability, it is said they are equivalent. The set of gauge transformations forms a group referred to as the gauge group. Since these transformations preserve the solution manifold it may also be called the symmetry group.

    Suppose the form $\omega (\lambda) = M (\lambda) dx + N (\lambda) \, dt$ is expanded in powers of $\lambda$ in the form $ \omega (\lambda) = \sum_{i=0}^m \, \omega_i \lambda^i, $ then if $S$ is an $n \times n$ matrix independent of $\lambda$, $\omega (S)$ is defined to be

    $ \omega (S) = \sum\limits_{i=0}^m \, \omega_i \, S^i. $ (4.13)

    Theorem 6. If there exists a gauge transformation of the form $\tau = \lambda I -S$ for the nonlinear evolution equation whose Lax representation is (2.1), then an $n \times n$ matrix $S$ independent of the parameter $\lambda$ must satisfy the following equation

    $ d S + [S, \omega (S)] =0. $ (4.14)

    Proof: Substitute the forms $\omega (\lambda) = \sum_{i=0}^m \, \omega_i \lambda^i$ and $\tilde{\omega} (\lambda) = \sum_{i=0}^m \, \tilde{\omega}_i \lambda^i$ and $\tau = \lambda \, I -S$ into $\Theta$ from (4.10). Collecting coefficients of the same powers of the spectral parameter $\lambda$

    $ - d S - \sum\limits_{i=1}^m \, \tilde{\omega}_i \lambda^{i+1} + \sum\limits_{i=1}^m \, \tilde{\omega}_i \lambda^i \, S + \sum\limits_{i=0}^m \, \omega_i \lambda^{i+1} - S \sum\limits_{i=0}^m \, \omega_i \lambda^i =0. $ (4.15)

    Collecting like powers of $\lambda$ in (4.15), there results,

    $ - d S + \tilde{\omega}_0 S - S \, \omega_0 + \sum\limits_{i=1}^m \, (- \tilde{\omega}_{i-1} + \tilde{\omega}_i \, S - \omega_{i-1} - S \omega_i ) \, \lambda^i + ( - \tilde{\omega}_m + \omega_m ) \lambda^{m+1} =0. $ (4.16)

    Equating the coefficients of each power $\lambda^i$ to zero, then if $i$ refers to the power on $\lambda$, the following system of equations results,

    $ d S + S \omega_0 - \tilde{\omega}_0 S =0, \quad i=0, \quad -\tilde{\omega}_{i-1} + \tilde{\omega}_i S + \omega_{i-1} - S \omega_i =0, \quad i=1, \ldots, m, $
    $ \quad - \tilde{\omega}_m + \omega_m =0, \quad i=m+1. $ (4.17)

    Applying these equations recursively, we can write

    $ - S \omega_j S^{j-1} + \tilde{\omega}_j S^j = - S \omega_j S^{j-1} + ( \omega_j - S \omega_{j+1} + \tilde{\omega}_{j+1} S ) S^j $
    $ = - S \omega_j S^{j-1} + \omega_j S^j - S \omega_{j+1} S^j + \tilde{\omega}_{j+1} S^{j+1} = [\omega_j, S] S^{j-1} - S \omega_{j+1} S^j + \tilde{\omega}_{j+1} S^{j+1}. $ (4.18)

    Thus iterating (4.18), a relationship between $\omega_0$ and $\tilde{\omega}_0$ can be obtained,

    $ \tilde{\omega}_0 = \omega_0 + [\omega_1, S] - S \omega_2 S + \tilde{\omega}_2 S^2 = \omega_0 + [\omega_1, S] + ( \omega_2 S - S \omega_2 ) \, S + S \omega_3 S^2 + \tilde{\omega}_3 \, S^3 $
    $ = \omega_0 + \sum\limits_{k=1}^m \, [\omega_k, S] \, S^{k-1}. $ (4.19)

    Solving for $dS$ from the $i=0$ equation of (4.17) and substituting $\tilde{\omega}_0$ from (4.19), we get

    $ d S = \tilde{\omega}_0 S - S \, \omega_0 = \omega_0 S - S \omega_0 + \sum\limits_{k=1}^m \, [\omega_k, S] S^k =- [S, \omega_0] - \sum\limits_{k=1}^m \, [S, \omega_k] \, S^k $
    $ =- \sum\limits_{k=0}^m \, [S, \omega_k] \, S^k. $

    Thus the result is exactly equation (4.14).

    Substituting $\omega (S) = M (S) \, dx + N (S) \, dt$ into (4.14), it breaks up into two equations as follows,

    $ S_x + [S, M (S)] =0, \qquad S_t + [S, N (S)] =0. $ (4.20)

    Theorem 7. System (4.14) is completely integrable if and only if the following equation is satisfied,

    $ d \omega (S) - \omega (S) \wedge \omega (S) =0, $ (4.21)

    or equivalently, if and only if

    $ M_t (S) - N_x (S) + [M (S), N (S)] =0. $ (4.22)

    Proof: Define the Pfaff system $\xi $ to be

    $ \xi = d S + [S, \omega (S)] = d S + [S, M (S)] \, dx + [S, N(S)] \, dt =0. $ (4.23)

    Differentiating $\xi$

    $ d \xi = [d S, M (S)] \wedge dx + [S, M_t (S)-N_x (S)] \, dt \wedge dx + [dS, N (S)] \wedge dt. $ (4.24)

    Obtaining $d S$ from (4.23) and replacing it in (4.24) gives,

    $ d \xi = [\xi, M (S)] \wedge dx + [\xi, N (S)] \wedge dt + [S, M_t (S)-N_x (S)] \, dt \wedge dx $
    $ -[[S, N (S)], M (S)] \, dt \wedge dx - [[S, M (S)], N (S)] \, dx \wedge dt. $ (4.25)

    Using the Jacobi identity, equation (4.25) simplifies to the form

    $ d \xi = [\xi, M (S)] \wedge dx + [\xi, N(S)] \wedge dt + [S, M_t (S)- N_x (S) + [M (S), N(S)]] \, dt \wedge dx. $

    Therefore, by the Frobenius Theorem, the Pfaff system $\xi =0$ is completely integrable if and only if (4.22) holds.

    Let matrix $S$ be constructed in the following way. Suppose $\Lambda$ is an $ n \times n$ matrix formed by putting the $n$ complex parameters $\lambda_1, \ldots, \lambda_n$ on the main diagonal and zero everywhere else. Then $B$ is an $n \times n$ invertible matrix such that $B = \text{diag} (\psi_1, \ldots, \psi_n)$, where the $\psi_i$, $i=1, \ldots, n$ satisfy the equations

    $ d \psi_i = \omega ( \lambda_i ) \, \psi_i, \qquad i =1, \ldots, n. $ (4.26)

    Define the matrix $S$ to be

    $ S = B \Lambda \, B^{-1}. $ (4.27)

    Theorem 8. The matrix $S$ from (4.27) satisfies the following relation,

    $ d S^i = \omega (S) \, S^i - S^i \, \omega (S). $ (4.28)

    Proof: Differentiating $B$, there results

    $ d B = \text{diag} ( d \psi_1, \ldots, d \psi_n ) = \text{diag} ( \omega (\lambda_1) \psi_1, \ldots, \omega ( \lambda_n ) \psi_n ) $
    $ = \text{diag} ( \sum\limits_{i=1}^m \, \omega_i \lambda_1^i \psi_1, \ldots, \sum\limits_{i=0}^m \, \omega_i \lambda_n^i \psi_n ) = \sum\limits_{i=0}^m \, \omega_i B \Lambda^i. $ (4.29)

    This can be used to evaluate $d S$,

    $ dS = d ( B \Lambda B^{-1} ) = d B \Lambda B^{-1} + B \Lambda d B^{-1} = dB \Lambda B^{-1} - B \Lambda B^{-1} \cdot dB B^{-1} $
    $ = ( \sum\limits_{i=0}^m \, \omega_i B \Lambda^i ) \Lambda B^{-1} - (B \Lambda B^{-1}) ( \sum\limits_{i=0}^m \omega_i B \Lambda^i ) B^{-1} $
    $ = ( \sum\limits_{i=0}^m \, \omega_i B \Lambda^i B^{-1} )( B \Lambda B^{-1}) - ( B \Lambda B^{-1}) ( \sum\limits_{i=0}^m \, \omega_i B \Lambda^i B^{-1}) $
    $ = ( \sum\limits_{i=0}^m \, \omega_i S^i ) S - S ( \sum\limits_{i=0}^m \, \omega_i S^i ) = \omega(S) S - S \omega (S). $ (4.30)

    Differentiating one $S$ at a time sequentially in $S^i$, we obtain

    $ d S^i = d S \, S^{i-1} + S dS \, S^{i-2} + \ldots + S^{i-1} \, dS. $ (4.31)

    Substituting $dS$ from (4.30) into each of these terms one after the other, the following system is obtained,

    $ dSSi1=(ω(S)SSω(S))Si1=ω(S)SiSω(S)Si1,SdSSi2=S(ω(S)SSω(S))Si2=Sω(S)Si1S2ω(S)Si2,Si1dS=Si1(ω(S)SSω(S))=Si1ω(S)SSiω(S).
    $

    Adding these together vertically most terms cancel out, as the sum telescopes, and the result (4.28) follows.

    Theorem 9. Matrix $S$ satisfies the equation

    $ \sum\limits_{i=0}^m \, d \omega_i S^i - \sum\limits_{i=0}^m \, \omega_i \wedge \omega (S) S^i =0. $ (4.32)

    Proof: The compatibility condition for the system (2.4) is

    $ d \omega ( \lambda ) \psi - \omega ( \lambda ) \wedge d \psi =0. $

    The curvature can be expanded into a series in powers of the spectral parameter $\lambda$ as

    $ \Omega ( \lambda) = \sum\limits_{i=0}^m \, d \omega_i \lambda^i - \sum\limits_{i=0}^m \, \omega_i \lambda^i \wedge \omega ( \lambda) =0. $ (4.33)

    Replacing $\lambda$ by the spectral parameters $\lambda_1, \ldots, \lambda_n$ one after the other yields the system of equations,

    $ mi=0dωiλi1=mi=0ωiω(λ1)λi1=mi,j=0ωiωjλi+j1,mi=0dωiλin=mi=0ωiω(λn)λin=mi,j=0ωiωjλi+jn.
    $

    Multiply the $j$-th equation in this set by $\psi_j$, so if $B$ represents the $n \times n$ matrix defined above, this system of equations can be written in the following way,

    $ \sum\limits_{i=0}^m \, d \omega_i B \Lambda^i = \sum\limits_{i, j=1}^m \, \omega_i \wedge \omega_j B \Lambda^{i+j}. $ (4.34)

    Finally, multiply (4.34) from the right on both sides by $B^{-1}$ and use the definition of $S$ in (4.27), and the result follows.

    Theorem 10. Matrix $S$ in (4.27) satisfies the relation

    $ d \omega (S) - \omega (S) \wedge \omega (S) =0. $

    Proof: Differentiating (4.13) and using Theorem 8 to replace $d S^i$, we obtain

    $ d \omega (S) = d ( \sum\limits_{i=0}^m \, \omega_i S^i) = \sum\limits_{i=0}^m \, d \omega_i S^i - \sum\limits_{i=0}^m \omega_i \wedge d S^i = \sum\limits_{i=0}^m \, d \omega_i S^i - \sum\limits_{i=0}^m \, \omega_i \wedge ( \omega (S) S^i - S^i \omega (S)) $
    $ = \sum\limits_{i=0}^m \, \omega_i \wedge S^i \omega (S) + [\sum\limits_{i=0}^m \, d \omega_i S^i-\sum\limits_{i=0}^m \omega_i \wedge \omega (S) S^i] = \omega (S) \wedge \omega (S). $

    In the second line, (4.32) has been used to eliminate the second term in the square brackets to finish the proof.


    [1] Colagiuri S, Colagiuri R, Ward J (1998) National diabetes strategy and implementation plan. Canberra: Paragon Printers.
    [2] Pop-Busui R, Evans GW, Gerstein HC, et al. (2010) The ACCORD Study Group. Effects of cardiac autonomic dysfunction on mortality risk in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) Trial. Diab Care 33: 1578-84.
    [3] Spallone V, Ziegler D, Freeman R, et al. (2011) Cardiovascular autonomic neuropathy in diabetes: clinical impact, assessment, diagnosis, and management. Diab Metab Res Rev 27:639-53. doi: 10.1002/dmrr.1239
    [4] Jelinek HF, Imam HM, Al-Aubaidy H, et al. (2013) Association of cardiovascular risk using nonlinear heart rate variability measures with the Framingham risk score in a rural population. Front Physiol 4: 186.
    [5] Ziegler D (1994) Diabetic cardiovascular autonomic neuropathy: prognosis, diagnosis and treatment. Diab Metabol Rev 10: 339-83. doi: 10.1002/dmr.5610100403
    [6] Gerritsen J, Dekker JM, TenVoorde BJ, et al. (2001) Impaired autonomic function is associated with increased mortality, especially in subjects with diabetes, hypertension or a history of cardiovascular disease. Diab Care 24: 1793-8. doi: 10.2337/diacare.24.10.1793
    [7] Johnston SC, Easton JD (2003) Are patients with acutely recovered cerebral ischemia more unstable? Stroke 4: 24-46.
    [8] Ko SH, Kwon HS, Lee JM, et al. (2006) Cardiovascular autonomic neuropathy in patients with type 2 diabetes mellitus. J Korean Diab Assoc 30: 226-35. doi: 10.4093/jkda.2006.30.3.226
    [9] Agelink MW, Malessa R, Baumann B, et al. (2001) Standardized tests of heart rate variability: normal ranges obtained from 309 healthy humans, and effects of age, gender and heart rate. Clin Auton Res 11: 99-108. doi: 10.1007/BF02322053
    [10] Ewing DJ, Martyn CN, Young RJ, et al. (1985) The value of cardiovascular autonomic functions tests: 10 years experience in diabetes. Diab Care 8: 491-8. doi: 10.2337/diacare.8.5.491
    [11] Pumprla J, Howorka K, Groves D, et al. (2002) Functional assessment of HRV variability: physiological basis and practical applications. Int J Cardiol 84: 1-14. doi: 10.1016/S0167-5273(02)00057-8
    [12] Stern S, Sclarowsky S (2009) The ECG in diabetes mellitus. Circulation 120: 1633-6. doi: 10.1161/CIRCULATIONAHA.109.897496
    [13] Reilly RB, Lee TC (2010) Electrograms (ECG, EEG, EMG, EOG). Technol Heal Care 18:443-58.
    [14] Baumert M, Schlaich MP, Nalivaiko E, et al. (2011) Relation between QT interval variability and cardiac sympathetic activity in hypertension. Am J Physiol Heart Circ Physiol 300: H1412-7. doi: 10.1152/ajpheart.01184.2010
    [15] Kelarev AV, Dazeley R, Stranieri A, et al. (2012) Detection of CAN by ensemble classifiers based on ripple down rules. Lect Notes Artif Int 7457: 147-59.
    [16] Fang ZY, Prins JB, Marwick TH (2004) Diabetic cardiomyopathy: evidence, mechanisms, and therapeutic implications. Endocrinol Rev 25: 543-67. doi: 10.1210/er.2003-0012
    [17] Huda S, Jelinek HF, Ray B, et al. (2010) Exploring novel features and decision rules to identify cardiovascular autonomic neuropathy using a Hybrid of Wrapper-Filter based feature selection. Marusic S, Palaniswami M, Gubbi J, et al, editors. Intelligent sensors, sensor networks and information processing, ISSNIP 2010. Sydney: IEEE Press, 297-302.
    [18] Witten H, Frank E (2011) Data mining: practical machine learning tools and techniques with java implementations. 3Eds, Sydney: Morgan Kaufmann, 2011.
    [19] Hall M, Frank E, Holmes G, et al. (2009). The WEKA data mining software: an update. SIGKDD Explor 11(1): 10-8. doi: 10.1145/1656274.1656278
    [20] Freund Y, Mason L (1999) The alternating decision tree learning algorithm. Proceedings of the sixteenth international conference on machine learning, 124-33.
    [21] Kotsiantis SB (2007) Supervised machine learning: A review of classification techniques. Informatica 31: 249-68.
    [22] Kohavi R (1996) Scaling up the accuracy of Naive-Bayes classifiers: a decision-tree hybrid. Proceedings of the 2nd international conference on knowledge discovery and data mining,202-7.
    [23] Breiman L, Friedman JH, Olshen RA, et al. (1984) Classification and regression trees. California: Wadsworth International Group.
    [24] Dazeley R, Yearwood J, Kang B, et al. (2010) Consensus clustering and supervised classification for profiling phishing emails in internet commerce secu+rity. In: Kang BH, Richards D, editors. Knowledge management and acquisition for smart systems and services, PKAW 2010. Daegu: Springer Verlag, 235-46.
    [25] Yearwood J, Webb D, Ma L, et al. (2009) Applying clustering and ensemble clustering approaches to phishing profiling. Proceedings of the 8th Australasian data mining conference, AusDM 2009. Curr Res Prac Inf Technol 101: 25-34.
    [26] Kang B, Kelarev A, Sale A, et al. (2006) A new model for classifying DNA code inspired by neural networks and FSA. Advances in Knowledge Acquisition and Management, 19th Australian Joint Conference on Artificial Intelligence, AI06. Lect Notes Comp Sci 4303:187-98.
    [27] Breiman L (1996) Bagging predictors. Mach Learn 24: 123-40.
    [28] Freund Y, Schapire R (1996) Experiments with a new boosting algorithm. Proceedings of 13th International Conference of Machince Learning 148-56.
    [29] Webb G (2000) Multiboosting: A technique for combining boosting and wagging. Mach Learn 40: 159-96. doi: 10.1023/A:1007659514849
    [30] Wolpert D (1992) Stacked generalization. Neural Networks 5: 241-59. doi: 10.1016/S0893-6080(05)80023-1
    [31] Melville P, Mooney R (2005) Creating diversity in ensembles using artificial data. Inf Fusion6: 99-111.
    [32] Ting K, Witten I (1997) Stacking bagged and dagged models. Fourteenth International Conference Machine Learning, 367-75.
    [33] Seewald AK, Furnkranz J (2001) An evaluation of grading classifiers. Hoffmann F, Adams N, Fisher D, et al, editors. Advances in intelligent data analysis, IDA 2001. Heidelberg: Springer,115-24.
    [34] Kelarev A, Stranieri A, Yearwood J, et al. (2012) Empirical study of decision trees and ensemble classifiers for monitoring of diabetes patients in pervasive healthcare, 15th International Conference on Networked-Based Information System, NBiS-2012. Melbourne: CPS, 441-6.
    [35] Huehn J, Huellermeier E (2009) FURIA: An algorithm for unordered fuzzy rule induction. DMKD 19: 293-319.
    [36] Kelarev A, Stranieri A, Yearwood J, et al. (2012) Improving classifications for cardiac autonomic neuropathy using multi-level ensemble classifiers and feature selection based on random forest. In: Zhao Y, Li J, Kennedy PJ, et al, editors. Data mining and analytics, 11th Australasian Data Mining Conference, AusDM 2012. Sydney: CRPIT, 134: 93-102.
  • Reader Comments
  • © 2014 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(6996) PDF downloads(1125) Cited by(9)

Article outline

Figures and Tables

Figures(4)  /  Tables(2)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog